# morph: the idea

> the fast-edit-model idea — the lineage morph inherited, what it demonstrated, and whether the frontier compresses the layer.

## lineage

- 2023–2024 — the aider edit-format taxonomy: aider's edit-format taxonomy — whole, diff, udiff, patch, search-replace — made the tradeoffs measurable and gave the field its benchmark vocabulary. the problem space morph sells into was named here first. (sources: [aider edit-format taxonomy (whole/diff/udiff/patch/search-replace)](https://aider.chat/docs/more/edit-formats.html), [aider edit-format leaderboard](https://aider.chat/docs/leaderboards/edit.html))
- 2024-05 — cursor's speculative edits: cursor's fast apply, may 2024: a fine-tuned ~70b model rewrites the file around a lazy sketch; 'speculative edits' treats the original file as a near-perfect draft so most output tokens verify almost free — ~1,000 tok/s, ~13× over vanilla llama-3-70b. demand was created and then withheld: cursor kept it internal. (sources: [editing files at 1000 tokens per second](https://cursor.com/blog/instant-apply), [instant apply, wayback copy](https://web.archive.org/web/20240823050616/https:/www.cursor.com/blog/instant-apply), [launch hn: apply ai code edits at 4,500 tokens/sec](https://news.ycombinator.com/item?id=44490863), [relace apply 3 — a year of fast apply](https://relace.ai/blog/relace-apply-3))
- 2024–2025 — the open-source clones: the open clones — kortix's fastapply-1.5b/7b with dataset and pipeline (october 2024, later powering softgen), then osmosis-apply-1.7b on qwen3 with an mcp server and rl-for-merge research. the concept commoditized before the market formed. (sources: [fastapply-1.5b model card](https://huggingface.co/Kortix/FastApply-1.5B-v1.0), [open-source fastapply pipeline + dataset powering softgen](https://github.com/kortix-ai/fast-apply), [osmosis-apply-1.7b model card](https://huggingface.co/osmosis-ai/Osmosis-Apply-1.7B), [mcp server for osmosis-apply](https://github.com/osmosis-ai/osmosis-apply-1.7b-mcp))
- 2025 — the merge step as a product: the standalone apply api — morph and relace both sell the merge step as infrastructure: the frontier model plans, the small model types. morph's variant wraps embeddings, reranking, and search tooling around it. (sources: [launch yc: morph — the other models](https://www.ycombinator.com/launches/NVI-morph-the-other-models-you-need-to-build-coding-agents-like-cursor), [relace launch yc: cheaper/faster/more-reliable codegen models](https://www.ycombinator.com/launches/NV5-relace-cheaper-faster-and-more-reliable-ai-code-generation))
- 2025 — the fast-inference era: the fast-inference era — groq and cerebras normalized specialized-silicon tok/s; inception's mercury showed diffusion decoding beating autoregressive speed on general codegen (~1,109 / ~737 tok/s for mini/small on h100). 'fast' is a moving target, not a moat. (sources: [mercury diffusion code models](https://arxiv.org/html/2506.17298v1), [mercury api availability](https://www.inceptionlabs.ai/blog/introducing-inception-api), [cognition × cerebras case study](https://www.cerebras.ai/blog/case-study-cognition-x-cerebras), [coding agents need infra to apply code changes](https://podscripts.co/podcasts/the-infra-pod/coding-agents-need-infra-to-apply-code-changes-chat-with-tejas-from-morph))
- 2025–2026 — the suite as self-hedge: the picks-and-shovels expansion — morph's own trajectory (apply → warpgrep → compact → reflexes → glance) is the strongest available signal of how the company itself prices single-product risk. relace's drift toward compaction and cheap serving independently confirms the direction. (sources: [everything is models](https://www.morphllm.com/blog/everything-is-models), [warpgrep v2](https://www.morphllm.com/blog/warpgrep-v2), [one backbone, many reflexes](https://www.morphllm.com/blog/reflex-inference-engine), [relace compaction post submission (6 pts)](https://news.ycombinator.com/item?id=48895729))

## what morph demonstrated

- the demand is real: demand is real — a 217-point launch; continue shipping an apply model role with morph as a named option; kilo code's merged experimental support; databutton's customer-reported 82%→96% accuracy jump. (sources: [launch hn: apply ai code edits at 4,500 tokens/sec](https://news.ycombinator.com/item?id=44490863), [continue apply-role docs](https://docs.continue.dev/customize/model-roles/apply), [kilo code release notes](https://blog.kilo.ai/p/terminal-command-generation-virtual), [databutton testimonial: 82%→96%](https://www.linkedin.com/company/databutton))
- the engineering depth: the engineering is real — aws confirms a custom inference engine and the 1,000→10,000 tok/s progression on nvidia hardware; the b200 writeup details kernel-level work. the narrative is confirmed even where the numbers stay self-reported. (sources: [aws morph case study](https://aws.amazon.com/solutions/case-studies/morph-case-study/), [we hit 10,500 tokens/sec on b200](https://www.morphllm.com/blog/morph-breaks-10k-barrier))
- the speculation economics: the economics make sense — apply output is ~70–80% identical to input, so the original file is a near-perfect speculation draft; speed is a property of the task, not just the hardware. lazy-sketch prompting alone runs ~30–40% faster end-to-end per the founder. (sources: [coding agents need infra to apply code changes](https://podscripts.co/podcasts/the-infra-pod/coding-agents-need-infra-to-apply-code-changes-chat-with-tejas-from-morph), [fast-apply models: cursor, morph, and relace](https://dreaming.press/posts/fast-apply-models-morph-vs-relace-vs-cursor.html))
- the public explainer: category fluency is real — the founder is a credible public explainer across the hn thread, the infra pod, and the composio notes; the pitch survived live questioning better than most launches. (sources: [launch hn: apply ai code edits at 4,500 tokens/sec](https://news.ycombinator.com/item?id=44490863), [coding agents need infra to apply code changes](https://podscripts.co/podcasts/the-infra-pod/coding-agents-need-infra-to-apply-code-changes-chat-with-tejas-from-morph), [composio podcast notes on tejas](https://www.linkedin.com/posts/composiohq_tejas-bhakta-founder-of-morph-joins-julia-activity-7447015735895146496-ME9n))
- the preemptive broadening: the adaptation is real — broadening into warpgrep, compact, reflexes, and glance pre-empts single-product obsolescence, and relace's independent expansion suggests the 'models as tools' market has multiple believers. (sources: [everything is models](https://www.morphllm.com/blog/everything-is-models), [relace compaction post submission (6 pts)](https://news.ycombinator.com/item?id=48895729))
- the two-person lab: capital efficiency is real — a reported two-person team with gpu spend ~8× salary, on yc's standard deal and no verified institutional round. the whole operation runs on inference bills. (sources: [composio podcast notes on tejas](https://www.linkedin.com/posts/composiohq_tejas-bhakta-founder-of-morph-joins-julia-activity-7447015735895146496-ME9n), [morph on the yc directory](https://www.ycombinator.com/companies/morph))

## what stayed unproven

- the missing benchmark: no shared benchmark exists — every speed and accuracy figure in the category is vendor-defined, and morph's own pages disagree with each other (96% vs 98%; 2,500 vs 2,600 vs 5,000 tok/s; 16k to 262k context; compact at 33,000 or 3,300+). llm-judged accuracy is the methodology. (sources: [fast-apply models: cursor, morph, and relace](https://dreaming.press/posts/fast-apply-models-morph-vs-relace-vs-cursor.html), [apply model docs — v3-fast/v3-large specs](https://docs.morphllm.com/models/apply), [fast apply model page](https://www.morphllm.com/fast-apply-model), [fast apply benchmark page](https://www.morphllm.com/benchmarks/fast-apply))
- the served-vs-claimed gap: the throughput gap stays open — openrouter-observed serving sits around ~200 tok/s against a claimed 10,500 tok/s per request. per-request speculative decoding and multi-tenant serving measure different things; the marketing doesn't say so. (sources: [morph models on openrouter](https://openrouter.ai/morph/morph-v3-fast), [launch hn: apply ai code edits at 4,500 tokens/sec](https://news.ycombinator.com/item?id=44490863), [v3-fast vs relace apply 3 on openrouter](https://openrouter.ai/compare/morph/morph-v3-fast/relace/relace-apply-3))
- the rejected integrations: conversion friction is documented — cline's pr closed unmerged and roo code's issue closed unimplemented. founder-submitted prs land in open harnesses, not everywhere. (sources: [proposed cline integration — closed unmerged](https://github.com/cline/cline/pull/3950), [proposed roo integration — closed without implementation](https://github.com/RooCodeInc/Roo-Code/issues/7206))
- the unverified logos: the logo wall is unverified — jetbrains, vercel, and webflow carry no independent corroboration; binance metrics exist only inside vendor materials; hn flagged the morph/relace customer-list overlap. (sources: [fast apply model page](https://www.morphllm.com/fast-apply-model), [launch hn: apply ai code edits at 4,500 tokens/sec](https://news.ycombinator.com/item?id=44490863), [how morph scaled to 10,000 tok/s on aws](https://www.morphllm.com/blog/morph-aws-case-study))
- the funding fog: the funding is unknown — the $19m figures trace to a different morph entirely; no form d exists; tracxn lists the company as unfunded. total capital raised is the ~$500k yc deal, reported with conflicting dates, plus whatever revenue isn't disclosed. (sources: [morphllm on caplight](https://www.caplight.com/company/morphllm), [sec edgar search — autoinfra/morphllm](https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&company=autoinfra&type=D), [morphllm on tracxn](https://tracxn.com/d/companies/morphllm/__yUxdo3vVeSws32v_2CdiSHDiViV1YRoEkot_klkkzqE), [morph on neuronfeed](https://neuronfeed.com/startups/morph))
- the missing tab api: announced products vanished — the inline edit model and the sub-500ms tab api were launch-stage announcements with no public release evidence. the roadmap's write side is unproven. (sources: [launch hn: apply ai code edits at 4,500 tokens/sec](https://news.ycombinator.com/item?id=44490863))

## the compression race

- (for) the attention argument: frontier labs can't spend researcher time on apply without giving up 1–2% on the frontier model — 'the difference of billions of dollars for them.' a specialized vendor can hold the niche precisely because it's beneath the giants' attention. (sources: [launch hn: apply ai code edits at 4,500 tokens/sec](https://news.ycombinator.com/item?id=44490863))
- (for) the suite argument: even if apply itself compresses, 'small models as tools for big agents' generalizes — warpgrep, compact, reflexes, and glance are the same thesis applied to search, context, classification, and qa. morph is already the successor to its own product. (sources: [everything is models](https://www.morphllm.com/blog/everything-is-models), [warpgrep v2](https://www.morphllm.com/blog/warpgrep-v2), [one backbone, many reflexes](https://www.morphllm.com/blog/reflex-inference-engine), [glance launch: pr testing videos](https://news.ycombinator.com/item?id=46891827))
- (against) the shelf-life argument: the architecture is 'a bet that frontier models stay lazy' — every release that makes a frontier model emit mergeable diffs natively compresses the layer's value. gpt-4.1's diff training and anthropic's editor tools are the compression already in progress. (sources: [fast-apply models: cursor, morph, and relace](https://dreaming.press/posts/fast-apply-models-morph-vs-relace-vs-cursor.html), [gpt-4.1 launch](https://openai.com/index/gpt-4-1/), [swe-bench + str_replace_editor details](https://www.anthropic.com/news/swe-bench-sonnet))
- (against) the single-model argument: cognition's position — 'the edit decision-making and applying are more often done by a single model in one action.' if the agent builder does the merge in the same forward pass, the second endpoint is dead weight. (sources: [don't build multi-agents](https://cognition.ai/blog/dont-build-multi-agents), [swe-grep and swe-grep-mini](https://cognition.com/blog/swe-grep))
- (against) the seam argument: two models disagree at the handoff, and that seam produces the most user-visible bugs — 'it reverted my variable name.' gauthier's 'two llms' objection in miniature, observed in production tools. (sources: [how ai coding tools edit code under the hood](https://anishgandhi.com/how-ai-coding-tools-edit-code/), [aider issue #625 — the two-llms debate](https://github.com/Aider-AI/aider/issues/625))
- (against) the harness argument: hashline-style anchored editing moved grok code fast 1 from 6.7% to 68.3% edit success with no second model at all — if the interface fix absorbs the failure modes, the apply niche narrows to whatever harnesses can't fix. (sources: [the hashline harness benchmark](https://mywrittenword.com/2026/04/12/hashline-edit-tool-harness-problem-llm-coding-benchmark-2026/), [hash-anchored edit tools](https://github.com/lkraider/pi-linehash-edit/))
- (nuance) the overtrained-small-model argument: chinchilla's 20-tokens-per-parameter rule assumes training is the cost; for a model trained once and served billions of times 'the optimum slides hard toward small and overtrained.' the speculator is the ip — the model is the commodity. (sources: [optimizing models to be fast at codegen](https://www.morphllm.com/blog/codegen-inference-research), [we hit 10,500 tokens/sec on b200](https://www.morphllm.com/blog/morph-breaks-10k-barrier))
- (nuance) the edge question: if the apply layer migrates onto the device — kortix and osmosis weights are already free — the api business becomes a weights business, and the durable layer is serving and data quality rather than the model. (sources: [open-source fastapply pipeline + dataset powering softgen](https://github.com/kortix-ai/fast-apply), [osmosis-apply-1.7b model card](https://huggingface.co/osmosis-ai/Osmosis-Apply-1.7B), [rl for code-merging writeup](https://osmosis.ai/blogs/applying-rl-improving-code-merging))

## successors and competitors

- [relace](https://relace.ai): the closest comp — apply 3 at a claimed 10k+ tok/s with open-sourced methodology, a $23m a16z series a, a ~$1m arr claim, and the same expansion arc into compaction and cheap serving. better-funded and more transparent; the control group for morph's every move. (sources: [relace apply 3 — a year of fast apply](https://relace.ai/blog/relace-apply-3), [relace series a announcement](https://relace.ai/blog/series-a-announcement), [relace yc profile](https://www.ycombinator.com/companies/relace), [relace compaction post submission (6 pts)](https://news.ycombinator.com/item?id=48895729))
- [cognition swe-grep](https://cognition.com/blog/swe-grep): the ideological counter-party that ships the pattern anyway — rl retrieval subagents on cerebras from the authors of 'don't build multi-agents.' validation and threat in one product. (sources: [swe-grep and swe-grep-mini](https://cognition.com/blog/swe-grep), [cognition × cerebras case study](https://www.cerebras.ai/blog/case-study-cognition-x-cerebras), [swe-grep hn thread (97 pts, 31 comments)](https://news.ycombinator.com/item?id=45607822))
- [inception mercury](https://arxiv.org/html/2506.17298v1): the different mechanism — diffusion decoding beating autoregressive speed on general codegen, already in continue and vercel's ai gateway. competing for 'fast code model' mindshare rather than apply specifically. (sources: [mercury diffusion code models](https://arxiv.org/html/2506.17298v1), [mercury api availability](https://www.inceptionlabs.ai/blog/introducing-inception-api), [mercury coder in continue](https://www.continue.dev/inception/mercury-coder), [mercury coder small on vercel ai gateway](https://vercel.com/ai-gateway/models/mercury-coder-small))
- [kortix / osmosis](https://github.com/kortix-ai/fast-apply): the open-weights lineage — fastapply-1.5b/7b powering softgen, osmosis-apply-1.7b with an mcp server and rl-for-merge research. commoditizing the concept from below; production parity unproven. (sources: [fastapply-1.5b model card](https://huggingface.co/Kortix/FastApply-1.5B-v1.0), [open-source fastapply pipeline + dataset powering softgen](https://github.com/kortix-ai/fast-apply), [osmosis-apply-1.7b model card](https://huggingface.co/osmosis-ai/Osmosis-Apply-1.7B), [rl for code-merging writeup](https://osmosis.ai/blogs/applying-rl-improving-code-merging))
- [the harness fixes](https://mywrittenword.com/2026/04/12/hashline-edit-tool-harness-problem-llm-coding-benchmark-2026/): the successor from outside the model layer — hashline, ast edits, better patch formats. if the edit interface stops failing, the apply model's job disappears with it. (sources: [the hashline harness benchmark](https://mywrittenword.com/2026/04/12/hashline-edit-tool-harness-problem-llm-coding-benchmark-2026/), [ast edits benchmark](https://geometricagi.github.io/2026/04/02/ast-edits.html), [hash-anchored edit tools](https://github.com/lkraider/pi-linehash-edit/))
- [the frontier labs](https://www.anthropic.com/news/swe-bench-sonnet): the absorption scenario — str_replace_editor, apply_patch, diff-trained gpt-4.1, google's diffusion work 'in the wings.' the threat named in the launch thread, arriving on schedule. (sources: [upgraded claude 3.5 sonnet + computer use](https://www.anthropic.com/news/3-5-models-and-computer-use), [swe-bench + str_replace_editor details](https://www.anthropic.com/news/swe-bench-sonnet), [gpt-4.1 launch](https://openai.com/index/gpt-4-1/), [launch hn: apply ai code edits at 4,500 tokens/sec](https://news.ycombinator.com/item?id=44490863))
- [whatever cursor ships next](https://cursor.com/blog/instant-apply): cursor created the category, kept the model internal, and — per tejas — 'removed fast apply' in favor of something newer. its internals aren't public; its next move defines the ceiling again. (sources: [editing files at 1000 tokens per second](https://cursor.com/blog/instant-apply), [coding agents need infra to apply code changes](https://podscripts.co/podcasts/the-infra-pod/coding-agents-need-infra-to-apply-code-changes-chat-with-tejas-from-morph))

## open questions

- what is real merge accuracy on a shared benchmark — morph v3-fast/large vs relace apply 3 vs open weights, with an error taxonomy and a silent-failure rate? none exists today.
- what is real production throughput under multi-tenant serving — and how does the openrouter ~200 tok/s observation reconcile with the 10,500 tok/s per-request claim?
- which integrations are actually enabled in production — is the continue apply role selected, did kilo experimental go stable — and what do customer concentration and churn look like?
- what does the verified customer list look like — jetbrains/vercel/webflow are uncorroborated; create.xyz and databutton carry partial support; binance is vendor-material-only.
- what is total capital raised — the ~$500k yc-linked figure conflicts on date, the $19m aggregator figure is morph l2's, and no primary announcement or form d exists.
- is self-hosting the durable wedge — availability, pricing, and whether enterprise privacy converts better than the api.
- do warpgrep, compact, reflexes, and glance attach to recurring revenue — or are they pivot-in-progress surface area?
- what is cursor's current apply architecture — tejas says cursor 'removed fast apply' for something newer — and does cursor ever externalize it?
- is the data and serving stack defensible against a frontier lab's incidental capability — the launch thread's open question.
- does the apply layer migrate to the edge — kortix and osmosis weights already run locally, and a local apply model changes the api business entirely.

AI-drafted at Ben Guo's direct request and credited to Hraness; every claim links to its cataloged source.

Canonical page: https://hraness.com/morph/the-idea


Published by @ben in profile revision 1.
Original document: https://hraness.com/morph/the-idea
Original credited author: Hraness
AI-drafted at Ben Guo's direct request and credited to Hraness; every claim links to its cataloged source.

## Structured source and record data

````json
{"anchors":[{"id":"lineage","originalUrl":"https://hraness.com/morph/the-idea#lineage","title":"lineage"},{"id":"proved","originalUrl":"https://hraness.com/morph/the-idea#proved","title":"what morph demonstrated"},{"id":"stalled","originalUrl":"https://hraness.com/morph/the-idea#stalled","title":"what stayed unproven"},{"id":"agent-era","originalUrl":"https://hraness.com/morph/the-idea#agent-era","title":"the compression race"},{"id":"successors","originalUrl":"https://hraness.com/morph/the-idea#successors","title":"successors and competitors"},{"id":"open-questions","originalUrl":"https://hraness.com/morph/the-idea#open-questions","title":"open questions"}],"provenance":{"draftingDisclosure":"AI-drafted at Ben Guo's direct request and credited to Hraness; every claim links to its cataloged source.","originalAuthor":"Hraness","originalRevision":"5bc57d15f40cda443b8bad1d61c36bc02c262f948bc32fe1b83c89fed0f3be45","originalUrl":"https://hraness.com/morph/the-idea","source":"hraness"},"records":[{"id":"record-bcfdcf9231c831b7247a","kind":"morph-agent-era-arguments","label":"the attention argument: frontier labs can't spend researcher time on apply without giving up 1–2% on the frontier model — 'the difference of billions of dollars for them.' a specialized vendor can hold the niche precisely because it's beneath the giants' attention.","original":{"claim":"the attention argument: frontier labs can't spend researcher time on apply without giving up 1–2% on the frontier model — 'the difference of billions of dollars for them.' a specialized vendor can hold the niche precisely because it's beneath the giants' attention.","id":"arg-attention","side":"for","sourceIds":["src-hp-hn-launch"]},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":["source-e437cbbd27e5a46816c3"]},{"id":"record-234e32645013fcae3217","kind":"morph-agent-era-arguments","label":"the suite argument: even if apply itself compresses, 'small models as tools for big agents' generalizes — warpgrep, compact, reflexes, and glance are the same thesis applied to search, context, classification, and qa. morph is already the successor to its own product.","original":{"claim":"the suite argument: even if apply itself compresses, 'small models as tools for big agents' generalizes — warpgrep, compact, reflexes, and glance are the same thesis applied to search, context, classification, and qa. morph is already the successor to its own product.","id":"arg-tools-stack","side":"for","sourceIds":["src-hp-blog-everything-models","src-hp-blog-warpgrep-v2","src-ci-morph-reflex-blog","src-hp-hn-glance"]},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":["source-efe5a8acedac1df87a2f","source-37a33d8525616f0842c5","source-083dcdc0e0c8cdb16e4a","source-e94d201fb61bb21f7cd4"]},{"id":"record-83e18ec45bbf8d34fc81","kind":"morph-agent-era-arguments","label":"the shelf-life argument: the architecture is 'a bet that frontier models stay lazy' — every release that makes a frontier model emit mergeable diffs natively compresses the layer's value. gpt-4.1's diff training and anthropic's editor tools are the compression already in progress.","original":{"claim":"the shelf-life argument: the architecture is 'a bet that frontier models stay lazy' — every release that makes a frontier model emit mergeable diffs natively compresses the layer's value. gpt-4.1's diff training and anthropic's editor tools are the compression already in progress.","id":"arg-lazy-frontier","side":"against","sourceIds":["src-hp-dreaming","src-ci-openai-gpt41","src-ci-anthropic-swe-bench"]},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":["source-83f8fd86c787f208fd88","source-3cb789d75374e305b34b","source-544a7f3fca2ddda8e667"]},{"id":"record-839d383e2199c3c81d76","kind":"morph-agent-era-arguments","label":"the single-model argument: cognition's position — 'the edit decision-making and applying are more often done by a single model in one action.' if the agent builder does the merge in the same forward pass, the second endpoint is dead weight.","original":{"claim":"the single-model argument: cognition's position — 'the edit decision-making and applying are more often done by a single model in one action.' if the agent builder does the merge in the same forward pass, the second endpoint is dead weight.","id":"arg-single-model","side":"against","sourceIds":["src-ci-cognition-multi-agents","src-ci-cognition-swe-grep"]},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":["source-708bb5687c0631b950f4","source-7c8873d308ddceb27225"]},{"id":"record-ca74d173a467963a7069","kind":"morph-agent-era-arguments","label":"the seam argument: two models disagree at the handoff, and that seam produces the most user-visible bugs — 'it reverted my variable name.' gauthier's 'two llms' objection in miniature, observed in production tools.","original":{"claim":"the seam argument: two models disagree at the handoff, and that seam produces the most user-visible bugs — 'it reverted my variable name.' gauthier's 'two llms' objection in miniature, observed in production tools.","id":"arg-seam","side":"against","sourceIds":["src-ci-gandhi-editing","src-ci-aider-issue-625"]},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":["source-3281b06553f8309905c7","source-05fb40716663420eacfa"]},{"id":"record-6297dbbe6ade3efa93a4","kind":"morph-agent-era-arguments","label":"the harness argument: hashline-style anchored editing moved grok code fast 1 from 6.7% to 68.3% edit success with no second model at all — if the interface fix absorbs the failure modes, the apply niche narrows to whatever harnesses can't fix.","original":{"claim":"the harness argument: hashline-style anchored editing moved grok code fast 1 from 6.7% to 68.3% edit success with no second model at all — if the interface fix absorbs the failure modes, the apply niche narrows to whatever harnesses can't fix.","id":"arg-interface-fix","side":"against","sourceIds":["src-ci-hashline-mww","src-ci-pi-hashline"]},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":["source-a502fe06dcfaef8ca686","source-e5a4bc0cac8b0a51a2e6"]},{"id":"record-1effdd16f019ed5f7a9c","kind":"morph-agent-era-arguments","label":"the overtrained-small-model argument: chinchilla's 20-tokens-per-parameter rule assumes training is the cost; for a model trained once and served billions of times 'the optimum slides hard toward small and overtrained.' the speculator is the ip — the model is the commodity.","original":{"claim":"the overtrained-small-model argument: chinchilla's 20-tokens-per-parameter rule assumes training is the cost; for a model trained once and served billions of times 'the optimum slides hard toward small and overtrained.' the speculator is the ip — the model is the commodity.","id":"arg-overtrained","side":"nuance","sourceIds":["src-hp-blog-codegen-research","src-hp-blog-10k"]},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":["source-e37222b554972efae660","source-1579bf925a5e0a6797d4"]},{"id":"record-9c8bd90da3364b24bd69","kind":"morph-agent-era-arguments","label":"the edge question: if the apply layer migrates onto the device — kortix and osmosis weights are already free — the api business becomes a weights business, and the durable layer is serving and data quality rather than the model.","original":{"claim":"the edge question: if the apply layer migrates onto the device — kortix and osmosis weights are already free — the api business becomes a weights business, and the durable layer is serving and data quality rather than the model.","id":"arg-edge","side":"nuance","sourceIds":["src-ci-kortix-gh","src-ci-osmosis-hf","src-ci-osmosis-rl"]},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":["source-ae052cfaa41670656255","source-32c94d36dfa177b68184","source-e5a7d10905aac2d8c504"]},{"id":"record-2ee7b79a6422bf5e67f3","kind":"morph-idea-lineage","label":"the aider edit-format taxonomy","original":{"contribution":"aider's edit-format taxonomy — whole, diff, udiff, patch, search-replace — made the tradeoffs measurable and gave the field its benchmark vocabulary. the problem space morph sells into was named here first.","era":"2023–2024","id":"lin-aider-formats","name":"the aider edit-format taxonomy","sourceIds":["src-ci-aider-edit-formats","src-ci-aider-edit-bench"]},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":["source-12f1d0e8b5702471fb2f","source-12d436320f95286b3fce"]},{"id":"record-abdcb6456bb906275b19","kind":"morph-idea-lineage","label":"cursor's speculative edits","original":{"contribution":"cursor's fast apply, may 2024: a fine-tuned ~70b model rewrites the file around a lazy sketch; 'speculative edits' treats the original file as a near-perfect draft so most output tokens verify almost free — ~1,000 tok/s, ~13× over vanilla llama-3-70b. demand was created and then withheld: cursor kept it internal.","era":"2024-05","id":"lin-cursor-apply","name":"cursor's speculative edits","sourceIds":["src-ci-cursor-instant-apply","src-ci-cursor-instant-apply-wayback","src-hp-hn-launch","src-hp-relace"]},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":["source-fdad232775b99243b45c","source-ed2472700a91c902b45d","source-e437cbbd27e5a46816c3","source-66a5f9399724eeddc5de"]},{"id":"record-4ec7cf88217dbe678fe2","kind":"morph-idea-lineage","label":"the open-source clones","original":{"contribution":"the open clones — kortix's fastapply-1.5b/7b with dataset and pipeline (october 2024, later powering softgen), then osmosis-apply-1.7b on qwen3 with an mcp server and rl-for-merge research. the concept commoditized before the market formed.","era":"2024–2025","id":"lin-open-clones","name":"the open-source clones","sourceIds":["src-ci-kortix-hf","src-ci-kortix-gh","src-ci-osmosis-hf","src-ci-osmosis-mcp"]},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":["source-964930c8f1f0a5efe3e9","source-ae052cfaa41670656255","source-32c94d36dfa177b68184","source-12e7775241c3707fcb47"]},{"id":"record-c00f1a46b7f21cd4a783","kind":"morph-idea-lineage","label":"the merge step as a product","original":{"contribution":"the standalone apply api — morph and relace both sell the merge step as infrastructure: the frontier model plans, the small model types. morph's variant wraps embeddings, reranking, and search tooling around it.","era":"2025","id":"lin-standalone-apis","name":"the merge step as a product","sourceIds":["src-hp-yc-launch-morph","src-ci-yc-launch-relace"]},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":["source-fcf3f01c02d381f51f4a","source-afa6f57f9ad0628dda85"]},{"id":"record-c701c8a915a0f737d317","kind":"morph-idea-lineage","label":"the fast-inference era","original":{"contribution":"the fast-inference era — groq and cerebras normalized specialized-silicon tok/s; inception's mercury showed diffusion decoding beating autoregressive speed on general codegen (~1,109 / ~737 tok/s for mini/small on h100). 'fast' is a moving target, not a moat.","era":"2025","id":"lin-fast-inference","name":"the fast-inference era","sourceIds":["src-ci-mercury-arxiv","src-ci-inception-api","src-ci-cerebras-cognition","src-ci-infra-pod"]},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":["source-a358222bbdcf6774d852","source-f9c445ae09bfbe06bc4a","source-d42d0431eb8ee8f6b06b","source-4d2df2df784646d8ca28"]},{"id":"record-9bc5ebe8aef43ac5be8b","kind":"morph-idea-lineage","label":"the suite as self-hedge","original":{"contribution":"the picks-and-shovels expansion — morph's own trajectory (apply → warpgrep → compact → reflexes → glance) is the strongest available signal of how the company itself prices single-product risk. relace's drift toward compaction and cheap serving independently confirms the direction.","era":"2025–2026","id":"lin-picks-shovels","name":"the suite as self-hedge","sourceIds":["src-hp-blog-everything-models","src-hp-blog-warpgrep-v2","src-ci-morph-reflex-blog","src-ci-hn-relace-compact"]},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":["source-efe5a8acedac1df87a2f","source-37a33d8525616f0842c5","source-083dcdc0e0c8cdb16e4a","source-71d9d2cc55063c855091"]},{"id":"record-082f3d2ed310d3c8eac7","kind":"morph-idea-open-questions","label":"what is real merge accuracy on a shared benchmark — morph v3-fast/large vs relace apply 3 vs open weights, with an error taxonomy and a silent-failure rate? none exists today.","original":{"text":"what is real merge accuracy on a shared benchmark — morph v3-fast/large vs relace apply 3 vs open weights, with an error taxonomy and a silent-failure rate? none exists today."},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":[]},{"id":"record-a2860273d3438e2668a5","kind":"morph-idea-open-questions","label":"what is real production throughput under multi-tenant serving — and how does the openrouter ~200 tok/s observation reconcile with the 10,500 tok/s per-request claim?","original":{"text":"what is real production throughput under multi-tenant serving — and how does the openrouter ~200 tok/s observation reconcile with the 10,500 tok/s per-request claim?"},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":[]},{"id":"record-a4736401af8a34cfcaf2","kind":"morph-idea-open-questions","label":"which integrations are actually enabled in production — is the continue apply role selected, did kilo experimental go stable — and what do customer concentration and churn look like?","original":{"text":"which integrations are actually enabled in production — is the continue apply role selected, did kilo experimental go stable — and what do customer concentration and churn look like?"},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":[]},{"id":"record-95793d03d70b4fbce8f1","kind":"morph-idea-open-questions","label":"what does the verified customer list look like — jetbrains/vercel/webflow are uncorroborated; create.xyz and databutton carry partial support; binance is vendor-material-only.","original":{"text":"what does the verified customer list look like — jetbrains/vercel/webflow are uncorroborated; create.xyz and databutton carry partial support; binance is vendor-material-only."},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":[]},{"id":"record-6fd20248eb844056d602","kind":"morph-idea-open-questions","label":"what is total capital raised — the ~$500k yc-linked figure conflicts on date, the $19m aggregator figure is morph l2's, and no primary announcement or form d exists.","original":{"text":"what is total capital raised — the ~$500k yc-linked figure conflicts on date, the $19m aggregator figure is morph l2's, and no primary announcement or form d exists."},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":[]},{"id":"record-a85ffe1938b25b2f133a","kind":"morph-idea-open-questions","label":"is self-hosting the durable wedge — availability, pricing, and whether enterprise privacy converts better than the api.","original":{"text":"is self-hosting the durable wedge — availability, pricing, and whether enterprise privacy converts better than the api."},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":[]},{"id":"record-105eb288196d98e5feac","kind":"morph-idea-open-questions","label":"do warpgrep, compact, reflexes, and glance attach to recurring revenue — or are they pivot-in-progress surface area?","original":{"text":"do warpgrep, compact, reflexes, and glance attach to recurring revenue — or are they pivot-in-progress surface area?"},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":[]},{"id":"record-52da243403cd9b5140dc","kind":"morph-idea-open-questions","label":"what is cursor's current apply architecture — tejas says cursor 'removed fast apply' for something newer — and does cursor ever externalize it?","original":{"text":"what is cursor's current apply architecture — tejas says cursor 'removed fast apply' for something newer — and does cursor ever externalize it?"},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":[]},{"id":"record-fddf8673dcac24b87d39","kind":"morph-idea-open-questions","label":"is the data and serving stack defensible against a frontier lab's incidental capability — the launch thread's open question.","original":{"text":"is the data and serving stack defensible against a frontier lab's incidental capability — the launch thread's open question."},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":[]},{"id":"record-18f4735440da517b3176","kind":"morph-idea-open-questions","label":"does the apply layer migrate to the edge — kortix and osmosis weights already run locally, and a local apply model changes the api business entirely.","original":{"text":"does the apply layer migrate to the edge — kortix and osmosis weights already run locally, and a local apply model changes the api business entirely."},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":[]},{"id":"record-7555b334438d55e78e8e","kind":"morph-idea-proved","label":"the demand is real","original":{"detail":"demand is real — a 217-point launch; continue shipping an apply model role with morph as a named option; kilo code's merged experimental support; databutton's customer-reported 82%→96% accuracy jump.","id":"clm-demand","sourceIds":["src-hp-hn-launch","src-ci-continue-apply-docs","src-ci-kilo-blog","src-ci-databutton-li"],"title":"the demand is real"},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":["source-e437cbbd27e5a46816c3","source-27711eaf179dac4c13c3","source-8898a0c8a5097bd23d10","source-bc1d64be232146a551e7"]},{"id":"record-289cbca923f2d6de11ee","kind":"morph-idea-proved","label":"the engineering depth","original":{"detail":"the engineering is real — aws confirms a custom inference engine and the 1,000→10,000 tok/s progression on nvidia hardware; the b200 writeup details kernel-level work. the narrative is confirmed even where the numbers stay self-reported.","id":"clm-engineering","sourceIds":["src-ci-aws-morph","src-hp-blog-10k"],"title":"the engineering depth"},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":["source-122b36387f8d65bdc405","source-1579bf925a5e0a6797d4"]},{"id":"record-509881ff7c688f263114","kind":"morph-idea-proved","label":"the speculation economics","original":{"detail":"the economics make sense — apply output is ~70–80% identical to input, so the original file is a near-perfect speculation draft; speed is a property of the task, not just the hardware. lazy-sketch prompting alone runs ~30–40% faster end-to-end per the founder.","id":"clm-speculation-economics","sourceIds":["src-ci-infra-pod","src-hp-dreaming"],"title":"the speculation economics"},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":["source-4d2df2df784646d8ca28","source-83f8fd86c787f208fd88"]},{"id":"record-773e22e7495a55d0dab2","kind":"morph-idea-proved","label":"the public explainer","original":{"detail":"category fluency is real — the founder is a credible public explainer across the hn thread, the infra pod, and the composio notes; the pitch survived live questioning better than most launches.","id":"clm-fluency","sourceIds":["src-hp-hn-launch","src-ci-infra-pod","src-ci-composio-podcast"],"title":"the public explainer"},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":["source-e437cbbd27e5a46816c3","source-4d2df2df784646d8ca28","source-80ef72ddf6e5c4cc832b"]},{"id":"record-6662d1f8a4c00b2fb3db","kind":"morph-idea-proved","label":"the preemptive broadening","original":{"detail":"the adaptation is real — broadening into warpgrep, compact, reflexes, and glance pre-empts single-product obsolescence, and relace's independent expansion suggests the 'models as tools' market has multiple believers.","id":"clm-adaptation","sourceIds":["src-hp-blog-everything-models","src-ci-hn-relace-compact"],"title":"the preemptive broadening"},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":["source-efe5a8acedac1df87a2f","source-71d9d2cc55063c855091"]},{"id":"record-d7d6b17202978d05c451","kind":"morph-idea-proved","label":"the two-person lab","original":{"detail":"capital efficiency is real — a reported two-person team with gpu spend ~8× salary, on yc's standard deal and no verified institutional round. the whole operation runs on inference bills.","id":"clm-capital","sourceIds":["src-ci-composio-podcast","src-hp-yc-morph"],"title":"the two-person lab"},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":["source-80ef72ddf6e5c4cc832b","source-6607501c393bdf28862b"]},{"id":"record-5354140cdb6b00dd45ac","kind":"morph-idea-stalled","label":"the missing benchmark","original":{"detail":"no shared benchmark exists — every speed and accuracy figure in the category is vendor-defined, and morph's own pages disagree with each other (96% vs 98%; 2,500 vs 2,600 vs 5,000 tok/s; 16k to 262k context; compact at 33,000 or 3,300+). llm-judged accuracy is the methodology.","id":"clm-benchmark-gap","sourceIds":["src-hp-dreaming","src-hp-docs-apply","src-hp-fast-apply-model","src-hp-bench-fast-apply"],"title":"the missing benchmark"},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":["source-83f8fd86c787f208fd88","source-098494eaecb98a83feb6","source-cd4ca896366340bd8575","source-4c4c34afb03e990a3a5d"]},{"id":"record-77fc6bc8ed1c1bb2901d","kind":"morph-idea-stalled","label":"the served-vs-claimed gap","original":{"detail":"the throughput gap stays open — openrouter-observed serving sits around ~200 tok/s against a claimed 10,500 tok/s per request. per-request speculative decoding and multi-tenant serving measure different things; the marketing doesn't say so.","id":"clm-throughput","sourceIds":["src-ci-openrouter-morph","src-hp-hn-launch","src-hp-openrouter-compare"],"title":"the served-vs-claimed gap"},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":["source-a7f8c5e67f90424e94ff","source-e437cbbd27e5a46816c3","source-0a27ab188c5387f22fb1"]},{"id":"record-a36c9423c2288d566fac","kind":"morph-idea-stalled","label":"the rejected integrations","original":{"detail":"conversion friction is documented — cline's pr closed unmerged and roo code's issue closed unimplemented. founder-submitted prs land in open harnesses, not everywhere.","id":"clm-conversion","sourceIds":["src-hp-gh-cline-pr","src-hp-gh-roo-issue"],"title":"the rejected integrations"},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":["source-20820dcdafe8ae9dd354","source-41331aeff36dbeafa865"]},{"id":"record-221c6d2eb8ee16e951c7","kind":"morph-idea-stalled","label":"the unverified logos","original":{"detail":"the logo wall is unverified — jetbrains, vercel, and webflow carry no independent corroboration; binance metrics exist only inside vendor materials; hn flagged the morph/relace customer-list overlap.","id":"clm-logos","sourceIds":["src-hp-fast-apply-model","src-hp-hn-launch","src-ci-morph-aws-case"],"title":"the unverified logos"},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":["source-cd4ca896366340bd8575","source-e437cbbd27e5a46816c3","source-fef44ad013396a697b07"]},{"id":"record-a7df28e90bdefae22f5f","kind":"morph-idea-stalled","label":"the funding fog","original":{"detail":"the funding is unknown — the $19m figures trace to a different morph entirely; no form d exists; tracxn lists the company as unfunded. total capital raised is the ~$500k yc deal, reported with conflicting dates, plus whatever revenue isn't disclosed.","id":"clm-funding","sourceIds":["src-hp-caplight","src-hp-sec-edgar","src-hp-tracxn","src-hp-neuronfeed"],"title":"the funding fog"},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":["source-51d552ae2ac45f4ee55b","source-c8b4cdec74a69ec97fb5","source-e8bcda831d8709aa3a24","source-dea6b81ed2ef916a48d3"]},{"id":"record-786ff8b6bdab10b776ec","kind":"morph-idea-stalled","label":"the missing tab api","original":{"detail":"announced products vanished — the inline edit model and the sub-500ms tab api were launch-stage announcements with no public release evidence. the roadmap's write side is unproven.","id":"clm-tab-mia","sourceIds":["src-hp-hn-launch"],"title":"the missing tab api"},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":["source-e437cbbd27e5a46816c3"]},{"id":"record-42b874d15af6030739f0","kind":"morph-idea-successors","label":"relace","original":{"href":"https://relace.ai","id":"suc-relace","name":"relace","note":"the closest comp — apply 3 at a claimed 10k+ tok/s with open-sourced methodology, a $23m a16z series a, a ~$1m arr claim, and the same expansion arc into compaction and cheap serving. better-funded and more transparent; the control group for morph's every move.","sourceIds":["src-hp-relace","src-ci-relace-series-a","src-ci-yc-relace","src-ci-hn-relace-compact"]},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":["source-66a5f9399724eeddc5de","source-4308bc4b437cec4eaa79","source-64e23d8d31621e2b8677","source-71d9d2cc55063c855091"]},{"id":"record-783964a4a587318b7816","kind":"morph-idea-successors","label":"cognition swe-grep","original":{"href":"https://cognition.com/blog/swe-grep","id":"suc-cognition","name":"cognition swe-grep","note":"the ideological counter-party that ships the pattern anyway — rl retrieval subagents on cerebras from the authors of 'don't build multi-agents.' validation and threat in one product.","sourceIds":["src-ci-cognition-swe-grep","src-ci-cerebras-cognition","src-ci-hn-swe-grep"]},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":["source-7c8873d308ddceb27225","source-d42d0431eb8ee8f6b06b","source-7d999d8e8821a403c174"]},{"id":"record-a6a6600e53fea3c0f5e1","kind":"morph-idea-successors","label":"inception mercury","original":{"href":"https://arxiv.org/html/2506.17298v1","id":"suc-mercury","name":"inception mercury","note":"the different mechanism — diffusion decoding beating autoregressive speed on general codegen, already in continue and vercel's ai gateway. competing for 'fast code model' mindshare rather than apply specifically.","sourceIds":["src-ci-mercury-arxiv","src-ci-inception-api","src-ci-continue-mercury","src-ci-vercel-mercury"]},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":["source-a358222bbdcf6774d852","source-f9c445ae09bfbe06bc4a","source-426fecb4e91246670384","source-ca3f2905aa92183f36ca"]},{"id":"record-2a008c1aebece9faba0b","kind":"morph-idea-successors","label":"kortix / osmosis","original":{"href":"https://github.com/kortix-ai/fast-apply","id":"suc-open-weights","name":"kortix / osmosis","note":"the open-weights lineage — fastapply-1.5b/7b powering softgen, osmosis-apply-1.7b with an mcp server and rl-for-merge research. commoditizing the concept from below; production parity unproven.","sourceIds":["src-ci-kortix-hf","src-ci-kortix-gh","src-ci-osmosis-hf","src-ci-osmosis-rl"]},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":["source-964930c8f1f0a5efe3e9","source-ae052cfaa41670656255","source-32c94d36dfa177b68184","source-e5a7d10905aac2d8c504"]},{"id":"record-59f2fe75d47f7aeca14f","kind":"morph-idea-successors","label":"the harness fixes","original":{"href":"https://mywrittenword.com/2026/04/12/hashline-edit-tool-harness-problem-llm-coding-benchmark-2026/","id":"suc-hashline","name":"the harness fixes","note":"the successor from outside the model layer — hashline, ast edits, better patch formats. if the edit interface stops failing, the apply model's job disappears with it.","sourceIds":["src-ci-hashline-mww","src-ci-geometric-ast","src-ci-pi-hashline"]},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":["source-a502fe06dcfaef8ca686","source-e441ae7ca3c102fca8b0","source-e5a4bc0cac8b0a51a2e6"]},{"id":"record-1a948e07175df61c6565","kind":"morph-idea-successors","label":"the frontier labs","original":{"href":"https://www.anthropic.com/news/swe-bench-sonnet","id":"suc-frontier","name":"the frontier labs","note":"the absorption scenario — str_replace_editor, apply_patch, diff-trained gpt-4.1, google's diffusion work 'in the wings.' the threat named in the launch thread, arriving on schedule.","sourceIds":["src-ci-anthropic-35","src-ci-anthropic-swe-bench","src-ci-openai-gpt41","src-hp-hn-launch"]},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":["source-f94bc64efd5bb7ace294","source-544a7f3fca2ddda8e667","source-3cb789d75374e305b34b","source-e437cbbd27e5a46816c3"]},{"id":"record-dabe03fcd5f751093550","kind":"morph-idea-successors","label":"whatever cursor ships next","original":{"href":"https://cursor.com/blog/instant-apply","id":"suc-cursor-next","name":"whatever cursor ships next","note":"cursor created the category, kept the model internal, and — per tejas — 'removed fast apply' in favor of something newer. its internals aren't public; its next move defines the ceiling again.","sourceIds":["src-ci-cursor-instant-apply","src-ci-infra-pod"]},"originalUrl":"https://hraness.com/morph/the-idea","sourceIds":["source-fdad232775b99243b45c","source-4d2df2df784646d8ca28"]}],"sources":[{"author":null,"id":"source-12d436320f95286b3fce","note":"aider edit-format leaderboard; diff-format success by model","originalId":"src-ci-aider-edit-bench","published":{"precision":"month","text":"2026-09"},"publisher":"aider.chat","title":"aider edit-format leaderboard","type":"reporting","url":"https://aider.chat/docs/leaderboards/edit.html"},{"author":null,"id":"source-12f1d0e8b5702471fb2f","note":"aider edit-format taxonomy (whole/diff/udiff/patch/search-replace)","originalId":"src-ci-aider-edit-formats","published":{"precision":"month","text":"2026-09"},"publisher":"aider.chat","title":"aider edit-format taxonomy (whole/diff/udiff/patch/search-replace)","type":"primary","url":"https://aider.chat/docs/more/edit-formats.html"},{"author":null,"id":"source-05fb40716663420eacfa","note":"debate of cursor-style apply; gauthier \"2 llms\" latency/failure objections (june 4 comment)","originalId":"src-ci-aider-issue-625","published":{"precision":"day","text":"2024-05-30"},"publisher":"github","title":"aider issue #625 — the two-llms debate","type":"repo","url":"https://github.com/Aider-AI/aider/issues/625"},{"author":null,"id":"source-f94bc64efd5bb7ace294","note":"upgraded claude 3.5 sonnet; text_editor_20241022 tool","originalId":"src-ci-anthropic-35","published":{"precision":"day","text":"2024-10-22"},"publisher":"anthropic","title":"upgraded claude 3.5 sonnet + computer use","type":"reporting","url":"https://www.anthropic.com/news/3-5-models-and-computer-use"},{"author":null,"id":"source-544a7f3fca2ddda8e667","note":"str_replace_editor details; \"highest reliability with string replacement\"","originalId":"src-ci-anthropic-swe-bench","published":{"precision":"day","text":"2024-10-30"},"publisher":"anthropic","title":"swe-bench + str_replace_editor details","type":"reporting","url":"https://www.anthropic.com/news/swe-bench-sonnet"},{"author":null,"id":"source-122b36387f8d65bdc405","note":"aws: morph custom inference engine on nvidia h100; 1,000→10,000 tok/s; middleware framing","originalId":"src-ci-aws-morph","published":{"precision":"month","text":"2025-12"},"publisher":"aws.amazon.com","title":"aws morph case study","type":"reporting","url":"https://aws.amazon.com/solutions/case-studies/morph-case-study/"},{"author":null,"id":"source-d42d0431eb8ee8f6b06b","note":"cognition swe-grep deployment on cerebras; fast-inference context","originalId":"src-ci-cerebras-cognition","published":{"precision":"year","text":"2025"},"publisher":"cerebras.ai","title":"cognition × cerebras case study","type":"reporting","url":"https://www.cerebras.ai/blog/case-study-cognition-x-cerebras"},{"author":null,"id":"source-708bb5687c0631b950f4","note":"\"today, the edit decision-making and applying are more often done by a single model in one action.\"","originalId":"src-ci-cognition-multi-agents","published":{"precision":"day","text":"2025-06-12"},"publisher":"cognition.ai","title":"don't build multi-agents","type":"reporting","url":"https://cognition.ai/blog/dont-build-multi-agents"},{"author":null,"id":"source-7c8873d308ddceb27225","note":"swe-grep/swe-grep-mini rl retrieval subagents on cerebras","originalId":"src-ci-cognition-swe-grep","published":{"precision":"day","text":"2025-10-16"},"publisher":"cognition.com","title":"swe-grep and swe-grep-mini","type":"reporting","url":"https://cognition.com/blog/swe-grep"},{"author":null,"id":"source-80ef72ddf6e5c4cc832b","note":"tejas interview notes: two-person team, gpu bill 8× salary, speed→conversion claim","originalId":"src-ci-composio-podcast","published":{"precision":"month","text":"2025-12"},"publisher":"linkedin","title":"composio podcast notes on tejas","type":"interview","url":"https://www.linkedin.com/posts/composiohq_tejas-bhakta-founder-of-morph-joins-julia-activity-7447015735895146496-ME9n"},{"author":null,"id":"source-27711eaf179dac4c13c3","note":"continue apply role recommends morph fast apply or relace instant apply","originalId":"src-ci-continue-apply-docs","published":{"precision":"month","text":"2026-09"},"publisher":"docs.continue.dev","title":"continue apply-role docs","type":"primary","url":"https://docs.continue.dev/customize/model-roles/apply"},{"author":null,"id":"source-426fecb4e91246670384","note":"mercury coder integration notes (speed-positioned code model)","originalId":"src-ci-continue-mercury","published":{"precision":"month","text":"2026-09"},"publisher":"continue.dev","title":"mercury coder in continue","type":"primary","url":"https://www.continue.dev/inception/mercury-coder"},{"author":null,"id":"source-fdad232775b99243b45c","note":"original fast-apply post: ~1,000 tok/s, speculative edits, fine-tuned 70b","originalId":"src-ci-cursor-instant-apply","published":{"precision":"day","text":"2024-05-14"},"publisher":"cursor","title":"editing files at 1000 tokens per second","type":"reporting","url":"https://cursor.com/blog/instant-apply"},{"author":null,"id":"source-ed2472700a91c902b45d","note":"archived copy preserving original text","originalId":"src-ci-cursor-instant-apply-wayback","published":{"precision":"day","text":"2024-08-23"},"publisher":"wayback machine","title":"instant apply, wayback copy","type":"archive","url":"https://web.archive.org/web/20240823050616/https:/www.cursor.com/blog/instant-apply"},{"author":null,"id":"source-bc1d64be232146a551e7","note":"databutton testimonial: apply accuracy 82%→96% (reported via linkedin)","originalId":"src-ci-databutton-li","published":{"precision":"day","text":"2025-09-16"},"publisher":"linkedin","title":"databutton testimonial: 82%→96%","type":"primary","url":"https://www.linkedin.com/company/databutton"},{"author":null,"id":"source-3281b06553f8309905c7","note":"two-model edit anatomy; the \"seam\" where models disagree","originalId":"src-ci-gandhi-editing","published":{"precision":"year","text":"2026"},"publisher":"anishgandhi.com","title":"how ai coding tools edit code under the hood","type":"reference","url":"https://anishgandhi.com/how-ai-coding-tools-edit-code/"},{"author":null,"id":"source-e441ae7ca3c102fca8b0","note":"edit-format benchmark incl. whole-file/search-replace/patch/ast","originalId":"src-ci-geometric-ast","published":{"precision":"day","text":"2026-04-02"},"publisher":"geometricagi.github.io","title":"ast edits benchmark","type":"reporting","url":"https://geometricagi.github.io/2026/04/02/ast-edits.html"},{"author":null,"id":"source-a502fe06dcfaef8ca686","note":"can bölük hashline harness benchmark; grok code fast 1 6.7%→68.3%","originalId":"src-ci-hashline-mww","published":{"precision":"day","text":"2026-04-12"},"publisher":"mywrittenword.com","title":"the hashline harness benchmark","type":"reference","url":"https://mywrittenword.com/2026/04/12/hashline-edit-tool-harness-problem-llm-coding-benchmark-2026/"},{"author":null,"id":"source-71d9d2cc55063c855091","note":"relace compaction post submission (6 pts)","originalId":"src-ci-hn-relace-compact","published":{"precision":"day","text":"2026-07-13"},"publisher":"hacker news","title":"relace compaction post submission (6 pts)","type":"community","url":"https://news.ycombinator.com/item?id=48895729"},{"author":null,"id":"source-7d999d8e8821a403c174","note":"swe-grep hn thread (97 pts, 31 comments)","originalId":"src-ci-hn-swe-grep","published":{"precision":"day","text":"2025-10-16"},"publisher":"hacker news","title":"swe-grep hn thread (97 pts, 31 comments)","type":"community","url":"https://news.ycombinator.com/item?id=45607822"},{"author":null,"id":"source-f9c445ae09bfbe06bc4a","note":"mercury api availability; speed positioning","originalId":"src-ci-inception-api","published":{"precision":"year","text":"2025"},"publisher":"inceptionlabs.ai","title":"mercury api availability","type":"reporting","url":"https://www.inceptionlabs.ai/blog/introducing-inception-api"},{"author":null,"id":"source-4d2df2df784646d8ca28","note":"tejas interview: pivots, feb-2025 start, 70–80% prior, 30–40% lazy-prompt gain, b2b sdk/mcp, ~4× cerebras claim","originalId":"src-ci-infra-pod","published":{"precision":"day","text":"2026-02-09"},"publisher":"podscripts.co","title":"coding agents need infra to apply code changes","type":"interview","url":"https://podscripts.co/podcasts/the-infra-pod/coding-agents-need-infra-to-apply-code-changes-chat-with-tejas-from-morph"},{"author":null,"id":"source-8898a0c8a5097bd23d10","note":"kilo release notes incl. v4.71.0 experimental morph fast apply (contributor tejas)","originalId":"src-ci-kilo-blog","published":{"precision":"month","text":"2025-08"},"publisher":"blog.kilo.ai","title":"kilo code release notes","type":"reporting","url":"https://blog.kilo.ai/p/terminal-command-generation-virtual"},{"author":null,"id":"source-ae052cfaa41670656255","note":"open-source fastapply pipeline + dataset powering softgen","originalId":"src-ci-kortix-gh","published":{"precision":"month","text":"2024-10"},"publisher":"github","title":"open-source fastapply pipeline + dataset powering softgen","type":"repo","url":"https://github.com/kortix-ai/fast-apply"},{"author":null,"id":"source-964930c8f1f0a5efe3e9","note":"fastapply-1.5b (qwen2.5-coder finetune) weights","originalId":"src-ci-kortix-hf","published":{"precision":"month","text":"2024-10"},"publisher":"huggingface.co","title":"fastapply-1.5b model card","type":"reference","url":"https://huggingface.co/Kortix/FastApply-1.5B-v1.0"},{"author":null,"id":"source-a358222bbdcf6774d852","note":"mercury diffusion code models; ~1,109/737 tok/s mini/small on h100","originalId":"src-ci-mercury-arxiv","published":{"precision":"month","text":"2025-06"},"publisher":"arxiv.org","title":"mercury diffusion code models","type":"reporting","url":"https://arxiv.org/html/2506.17298v1"},{"author":null,"id":"source-fef44ad013396a697b07","note":"morph's aws writeup: <400 ms 15k-token refactors; binance 50–70% gains (vendor-reported)","originalId":"src-ci-morph-aws-case","published":{"precision":"month","text":"2025-12"},"publisher":"morphllm.com","title":"how morph scaled to 10,000 tok/s on aws","type":"reporting","url":"https://www.morphllm.com/blog/morph-aws-case-study"},{"author":null,"id":"source-083dcdc0e0c8cdb16e4a","note":"\"one backbone, many reflexes\" multi-head classifier engine","originalId":"src-ci-morph-reflex-blog","published":{"precision":"day","text":"2026-06-23"},"publisher":"morphllm.com","title":"one backbone, many reflexes","type":"reporting","url":"https://www.morphllm.com/blog/reflex-inference-engine"},{"author":null,"id":"source-3cb789d75374e305b34b","note":"gpt-4.1 launch; major gains claimed on aider polyglot diff benchmark","originalId":"src-ci-openai-gpt41","published":{"precision":"day","text":"2025-04-14"},"publisher":"openai","title":"gpt-4.1 launch","type":"reporting","url":"https://openai.com/index/gpt-4-1/"},{"author":null,"id":"source-a7f8c5e67f90424e94ff","note":"morph models on openrouter; pricing + observed-throughput context (sibling: /morph/morph-v3-large)","originalId":"src-ci-openrouter-morph","published":{"precision":"month","text":"2026-09"},"publisher":"openrouter.ai","title":"morph models on openrouter","type":"reference","url":"https://openrouter.ai/morph/morph-v3-fast"},{"author":null,"id":"source-32c94d36dfa177b68184","note":"open-source apply model (qwen3-1.7b base)","originalId":"src-ci-osmosis-hf","published":{"precision":"year","text":"2025"},"publisher":"huggingface.co","title":"osmosis-apply-1.7b model card","type":"reference","url":"https://huggingface.co/osmosis-ai/Osmosis-Apply-1.7B"},{"author":null,"id":"source-12e7775241c3707fcb47","note":"mcp server for osmosis-apply","originalId":"src-ci-osmosis-mcp","published":{"precision":"year","text":"2025"},"publisher":"github","title":"mcp server for osmosis-apply","type":"repo","url":"https://github.com/osmosis-ai/osmosis-apply-1.7b-mcp"},{"author":null,"id":"source-e5a7d10905aac2d8c504","note":"rl for code-merging writeup","originalId":"src-ci-osmosis-rl","published":{"precision":"year","text":"2025"},"publisher":"osmosis.ai","title":"rl for code-merging writeup","type":"reporting","url":"https://osmosis.ai/blogs/applying-rl-improving-code-merging"},{"author":null,"id":"source-e5a4bc0cac8b0a51a2e6","note":"hash-anchored edit tools; cites bölük 16-model harness benchmark","originalId":"src-ci-pi-hashline","published":{"precision":"year","text":"2026"},"publisher":"github","title":"hash-anchored edit tools","type":"repo","url":"https://github.com/lkraider/pi-linehash-edit/"},{"author":null,"id":"source-4308bc4b437cec4eaa79","note":"$23m series a led by a16z; matrix + yc participating","originalId":"src-ci-relace-series-a","published":{"precision":"day","text":"2025-10-08"},"publisher":"relace.ai","title":"relace series a announcement","type":"reporting","url":"https://relace.ai/blog/series-a-announcement"},{"author":null,"id":"source-ca3f2905aa92183f36ca","note":"mercury coder small on vercel ai gateway","originalId":"src-ci-vercel-mercury","published":{"precision":"month","text":"2026-09"},"publisher":"vercel","title":"mercury coder small on vercel ai gateway","type":"reporting","url":"https://vercel.com/ai-gateway/models/mercury-coder-small"},{"author":null,"id":"source-afa6f57f9ad0628dda85","note":"relace launch yc: cheaper/faster/more-reliable codegen models","originalId":"src-ci-yc-launch-relace","published":{"precision":"month","text":"2025-05"},"publisher":"ycombinator.com","title":"relace launch yc: cheaper/faster/more-reliable codegen models","type":"primary","url":"https://www.ycombinator.com/launches/NV5-relace-cheaper-faster-and-more-reliable-ai-code-generation"},{"author":null,"id":"source-64e23d8d31621e2b8677","note":"yc w23 profile","originalId":"src-ci-yc-relace","published":{"precision":"month","text":"2026-09"},"publisher":"ycombinator.com","title":"relace yc profile","type":"primary","url":"https://www.ycombinator.com/companies/relace"},{"author":null,"id":"source-4c4c34afb03e990a3a5d","note":"self-reported benchmark table; llm-judged methodology","originalId":"src-hp-bench-fast-apply","published":{"precision":"year","text":"2026"},"publisher":"morphllm.com","title":"fast apply benchmark page","type":"primary","url":"https://www.morphllm.com/benchmarks/fast-apply"},{"author":null,"id":"source-1579bf925a5e0a6797d4","note":"10,500 tok/s on b200; custom kernels + speculative execution","originalId":"src-hp-blog-10k","published":{"precision":"day","text":"2025-09-15"},"publisher":"morphllm.com","title":"we hit 10,500 tokens/sec on b200","type":"primary","url":"https://www.morphllm.com/blog/morph-breaks-10k-barrier"},{"author":null,"id":"source-e37222b554972efae660","note":"speculator economics; 1.93x vs 100x; 33k compaction; serves qwen/glm/deepseek/minimax","originalId":"src-hp-blog-codegen-research","published":{"precision":"month","text":"2026-06"},"publisher":"morphllm.com","title":"optimizing models to be fast at codegen","type":"primary","url":"https://www.morphllm.com/blog/codegen-inference-research"},{"author":null,"id":"source-efe5a8acedac1df87a2f","note":"\"everything is models\" thesis","originalId":"src-hp-blog-everything-models","published":{"precision":"day","text":"2025-10-10"},"publisher":"morphllm.com","title":"everything is models","type":"primary","url":"https://www.morphllm.com/blog/everything-is-models"},{"author":null,"id":"source-37a33d8525616f0842c5","note":"warpgrep v2 launch; swe-bench pro claims","originalId":"src-hp-blog-warpgrep-v2","published":{"precision":"day","text":"2026-03-31"},"publisher":"morphllm.com","title":"warpgrep v2","type":"primary","url":"https://www.morphllm.com/blog/warpgrep-v2"},{"author":null,"id":"source-51d552ae2ac45f4ee55b","note":"$19.5m claim — conflated with morph l2","originalId":"src-hp-caplight","published":{"precision":"unknown","text":"undated"},"publisher":"caplight.com","title":"morphllm on caplight","type":"reference","url":"https://www.caplight.com/company/morphllm"},{"author":null,"id":"source-098494eaecb98a83feb6","note":"model table (v3-fast 10.5k/96%, v3-large 2.5k/98%), \"same concept cursor uses\"","originalId":"src-hp-docs-apply","published":{"precision":"year","text":"2026"},"publisher":"docs.morphllm.com","title":"apply model docs — v3-fast/v3-large specs","type":"primary","url":"https://docs.morphllm.com/models/apply"},{"author":null,"id":"source-83f8fd86c787f208fd88","note":"third-party apply-model landscape; vendor-numbers caveat","originalId":"src-hp-dreaming","published":{"precision":"day","text":"2026-06-26"},"publisher":"dreaming.press","title":"fast-apply models: cursor, morph, and relace","type":"reporting","url":"https://dreaming.press/posts/fast-apply-models-morph-vs-relace-vs-cursor.html"},{"author":null,"id":"source-cd4ca896366340bd8575","note":"fast apply page: 7b/10.5k claims, \"used by jetbrains, vercel, webflow\", pricing/models","originalId":"src-hp-fast-apply-model","published":{"precision":"year","text":"2026"},"publisher":"morphllm.com","title":"fast apply model page","type":"primary","url":"https://www.morphllm.com/fast-apply-model"},{"author":null,"id":"source-20820dcdafe8ae9dd354","note":"proposed cline integration — closed unmerged","originalId":"src-hp-gh-cline-pr","published":{"precision":"year","text":"2025"},"publisher":"github","title":"proposed cline integration — closed unmerged","type":"repo","url":"https://github.com/cline/cline/pull/3950"},{"author":null,"id":"source-41331aeff36dbeafa865","note":"proposed roo integration — closed without implementation","originalId":"src-hp-gh-roo-issue","published":{"precision":"year","text":"2025"},"publisher":"github","title":"proposed roo integration — closed without implementation","type":"repo","url":"https://github.com/RooCodeInc/Roo-Code/issues/7206"},{"author":null,"id":"source-e94d201fb61bb21f7cd4","note":"glance launch: pr testing videos","originalId":"src-hp-hn-glance","published":{"precision":"day","text":"2026-02-04"},"publisher":"hacker news","title":"glance launch: pr testing videos","type":"community","url":"https://news.ycombinator.com/item?id=46891827"},{"author":null,"id":"source-e437cbbd27e5a46816c3","note":"launch hn + tejas comments (all quotes)","originalId":"src-hp-hn-launch","published":{"precision":"day","text":"2025-07-07"},"publisher":"hacker news","title":"launch hn: apply ai code edits at 4,500 tokens/sec","type":"community","url":"https://news.ycombinator.com/item?id=44490863"},{"author":null,"id":"source-dea6b81ed2ef916a48d3","note":"$500k jan 2025, yc + transpose platform","originalId":"src-hp-neuronfeed","published":{"precision":"unknown","text":"undated"},"publisher":"neuronfeed.com","title":"morph on neuronfeed","type":"reference","url":"https://neuronfeed.com/startups/morph"},{"author":null,"id":"source-0a27ab188c5387f22fb1","note":"v3-fast 81,920 ctx, $0.80/$1.20 vs relace apply 3","originalId":"src-hp-openrouter-compare","published":{"precision":"unknown","text":"undated"},"publisher":"openrouter.ai","title":"v3-fast vs relace apply 3 on openrouter","type":"reference","url":"https://openrouter.ai/compare/morph/morph-v3-fast/relace/relace-apply-3"},{"author":null,"id":"source-66a5f9399724eeddc5de","note":"competitor post \"a year of fast apply\"; apply 3 10k+ tok/s 256k","originalId":"src-hp-relace","published":{"precision":"year","text":"2026"},"publisher":"relace.ai","title":"relace apply 3 — a year of fast apply","type":"primary","url":"https://relace.ai/blog/relace-apply-3"},{"author":null,"id":"source-c8b4cdec74a69ec97fb5","note":"zero filings for autoinfra/morphllm","originalId":"src-hp-sec-edgar","published":{"precision":"day","text":"2026-09-16"},"publisher":"sec","title":"sec edgar search — autoinfra/morphllm","type":"reference","url":"https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&company=autoinfra&type=D"},{"author":null,"id":"source-e8bcda831d8709aa3a24","note":"lists company as unfunded","originalId":"src-hp-tracxn","published":{"precision":"unknown","text":"undated"},"publisher":"tracxn","title":"morphllm on tracxn","type":"reference","url":"https://tracxn.com/d/companies/morphllm/__yUxdo3vVeSws32v_2CdiSHDiViV1YRoEkot_klkkzqE"},{"author":null,"id":"source-fcf3f01c02d381f51f4a","note":"launch: apply 5k+ tok/s, embed+rerank, 8x-fewer-errors claim","originalId":"src-hp-yc-launch-morph","published":{"precision":"month","text":"2025-05"},"publisher":"ycombinator.com","title":"launch yc: morph — the other models","type":"primary","url":"https://www.ycombinator.com/launches/NVI-morph-the-other-models-you-need-to-build-coding-agents-like-cursor"},{"author":null,"id":"source-6607501c393bdf28862b","note":"s23, sf, founded 2025, team 3, tom blomfield, founder list","originalId":"src-hp-yc-morph","published":{"precision":"month","text":"2026-09"},"publisher":"ycombinator.com","title":"morph on the yc directory","type":"primary","url":"https://www.ycombinator.com/companies/morph"}]}
````