history
the company, sourced and dated — lingtual to innkeeper to autoinfra to morph, the hn launch, and the suite broadening.
the pivot years2023–2024
yc s23 as lingtual, the innkeeper detour, autoinfra on the footer — and the morph l2 collision that poisoned the funding record.
- 2023-06-01the $500k pre-seed listinglinkedin lists a $500k pre-seed dated june 1, 2023 — consistent with the standard yc s23 deal. neuronfeed lists the same amount for january 2025 with transpose platform as co-investor. the two records disagree on date and partner; the amount stays reported, unreconciled.
- 2023-07-10the github orgthe github org morphllm is created mid-batch — almost certainly under an earlier name and renamed. it now holds ~24 public repos.
- 2023-07yc s23, as lingtualthe yc summer 2023 batch, sf — as lingtual: 'the easiest way to build connected ai business software,' a no-code builder pitched as retool+zapier for ai applications with 50+ integrations. the same yc profile later carried aka names autoinfra and innkeeper.
- 2023-08-04lingtual.comlingtual.com registers mid-batch; tool directories index the product by late july under founders@lingtual.com.
- 2023-12-11the first misattributioncaplight lists a 'corporate funding round' for morphllm — almost certainly morph l2's angel money attached to the wrong morph. the first recorded conflation.
- 2024-02-13the backdated blog'faster agents' (02-13) and 'how cursor composer and apply work' (02-15) are the earliest dated posts on the morph blog — a year before morphllm.com existed. migrated or backdated; treat the dates as approximate.
- 2024-03-20the other morph raises $19mmorph l2 — a different company, an ethereum layer 2 — announces a $19m dragonfly-led seed; fortune covers it. every '$19m raised' claim attached to morphllm traces here.
- 2024the innkeeper erathe innkeeper era — hotel and motel software per yc's aka list and third-party trackers. the pivot that matches bhakta's motel upbringing; ordering against autoinfra is unverified.
the morph rebrand2025-02–2025-06
morphllm.com registers in february; the apply model gets a launch post, a merged continue pr, and a cline pr that won't land.
- 2025-02-06morphllm.commorphllm.com registers — the current brand begins. the legal entity stays autoinfra, inc., per the october-2025 footer.
- 2025-02tejas starts morphtejas starts morph after three or four yc pivots, per his infra pod account — the yc directory's 'founded: 2025' field dates the product, not the entity.
- 2025-04-29the continue prtejas-authored continue pr #4663 merges, adding apply-role docs that name morph — public documentation ahead of the public launch.
- 2025-05-02the best-practices post'best practices for building coding agents with morph' — the blog turns toward agent-builders.
- 2025-05-11the self-host offer'self hosting' posts the self-serve deployment offer — the models for private infrastructure, before the launch.
- 2025-05-13launch yc: the 'other models'launch yc: 'the "other models" you need to build coding agents like cursor' — apply plus embedding and rerank models behind an openai-compatible api, and an '8x fewer errors than patch-based edits' claim attributed to an unnamed fortune-50 staff engineer. the launch copy reads 1,600+ tok/s; the current page reads 5,000+ with ~2s on a 9,000-token file — vendor figures drift across versions.
- 2025-05-28the model-vs-format argument'diffs vs fast apply' makes the model-over-format argument; 'morph ❤️ continue' claims the integration the same day.
- 2025-05-31the cline prtejas opens cline pr #3950 — an edit_file tool wired to the morph api with provider settings and api-key handling. closed unmerged on 2025-07-26.
- 2025-06-15'morph gets faster''morph gets faster' posts — the archived index later showed the title as '10,500+ tokens per second,' so the page was likely updated after september. the ambiguity is preserved.
- 2025-06-26morph-v2 on openroutermorph-v2 lists on openrouter — the first public model listing, days before the hn launch.
the launch2025-07–2025-09
the 217-point hn launch, the mcp server, kilo code's merge, and the 10,500-tok/s claim.
- 2025-07-07launch hn: 217 pointslaunch hn: 217 points, 185 comments. morph-v3-fast at a claimed 4,500+ tok/s, v3-large 2,500+, plus embedding and rerank models; the inline edit model is announced and the tab api (sub-500ms next-edit) enters private beta. the thread splits on bundling, bottlenecks, and privacy — and a playground bug becomes a small trust incident.
- 2025-07-18the mcp servershow hn: 'mcp for fast ai filesystem edits' — the mcp server ships. docs list setup targets across cursor, claude code, windsurf, cline, vs code, and claude desktop — setup targets, not partnerships.
- 2025-07-31kilo code mergeskilo code pr #1628 merges experimental morph fast apply — the first verified third-party shipped integration; a follow-up pr adds the badge, and the edit-file prompts support modeltype 'morph' | 'relace'.
- 2025roo code says noroo code issue #7206 proposes optional morph fast apply — approved in principle, then closed without implementation. proposal only.
- 2025-08-22the explainer'what is morph fast apply?' — the explainer post.
- 2025-08-26the fly.io tutorialfly.io publishes 'build better agents with morphllm' on dev.to — the first third-party integration walkthrough, mirrored at engineering.fyi.
- 2025-09-10'lol, i never said that'tejas replies to nik pash's 'fast apply models are already dead': 'lol, i never said that' — disputing the claimed six-month shelf-life concession and citing hundreds of iterations to reach 98% accuracy.
- 2025-09-1510,500 tok/s on b200'we hit 10,500 tokens/sec on b200' — custom cuda kernels plus a specialized speculator; 'roughly 4× faster than the best speeds on cerebras,' and under four seconds for complex edits across a 40k-token document. the hn submission draws 2 points.
- 2025-09-16the first customer metricdatabutton posts the only named customer metric in the record — apply accuracy 82%→96%, reported via linkedin.
the model suite2025-10–2026-02
'everything is models,' the aws case study, warpgrep, and glance — one model becomes a stack.
- 2025-10'everything is models''everything is models' pairs with 'llms are bad at being forced' — the thesis generalizes from one apply model to many small specialized models as tools for big agents. the hn submission draws 1 point. (the sources disagree on the day — 10-06 vs 10-10; the month is what holds.)
- 2025-10-11the october snapshotthe archived homepage reads 'small models as tools for big agents': a claimed 10,500 tok/s at 98% first-pass accuracy, full self-reported benchmark tables, 'trusted by hundreds of startups and enterprises.' the about footer still says autoinfra, inc.
- 2025-12the aws case studyaws publishes a morph case study — a custom inference engine on nvidia h100s and the 1,000→10,000 tok/s progression, framed as 'middleware' for frontier coding solutions. morph's own writeup adds under-400ms 15k-token refactors and binance 50–70% productivity claims — reported through vendor materials only.
- 2025-12-09warpgrepshow hn: warpgrep — an rl code-search subagent 'like swe-grep,' the first non-apply model. 1 point.
- 2026-01warpgrep on launch yclaunch yc: 'warpgrep — a 20x faster subagent to grep for code': up to 36 parallel calls under five seconds, swe-bench-pro gains, ~17% fewer input tokens — self-reported, and the '178s saved' arithmetic doesn't check out.
- 2026-01-20the twitter-algorithm demoshow hn: parallel agentic search on the open-sourced twitter recommendation repo — the warpgrep demo. 13 points.
- 2026-01-23learning to test code changesa show hn pair — 'learning to test code changes with rl' and agentic browser-testing videos in prs — plus the browser-verification post. the glance preview.
- 2026-02-04glanceshow hn: glance launches — videos of ai testing your pr, embedded in github, powered by morph-computer-use-v0. 35 points.
- 2026-02-09the infra podthe infra pod interview: the pivots, the february-2025 start, the 70–80%-copy figure, the ~4×-cerebras claim, and the b2b sdk/mcp strategy.
the subagent era2026-03–
warpgrep v2's swe-bench claims, the codegen research, and reflexes — the subagent era.
- 2026-03-27the start podcastthe start podcast: 'agents don't need bigger models. they need better tools.' — apply ~10k tok/s, compact 33k, 'morph trains coding subagents. not for humans. for frontier models.'
- 2026-03-31warpgrep v2warpgrep v2 launches on yc and the blog — '#1 on swe-bench pro'-style retrieval claims, ~900 tok/s on b200 vs swe-grep's ~650 on cerebras. the blog post ran earlier — dated 2026-02-23 in the coverage lane's record; the date conflict is preserved.
- 2026-06the codegen research'optimizing models to be fast at codegen': 'a generic draft gets 1.93x, a specilized one got 100x faster'; the overtrained-small-model argument ('the optimum slides hard toward small and overtrained'); and serving qwen, glm, deepseek, and minimax for agents. hn: 2 points.
- 2026-06-30reflexesshow hn: reflexes — multi-head classifiers on agent traces, a claimed <30ms inference; 'four or 100 reflexes add less than 2ms.' the companion post 'one backbone, many reflexes' ran 06-23.
- 2026-09the suite, as of researchthe suite as of research: fast apply, warpgrep, compact/flashcompact, reflexes, glance, morph-embedding-v4, morph-rerank-v4, an auto router, and genkit — plus first-party plugins for opencode, claude code, pi, and openclaw.
AI-drafted at Ben Guo's direct request and credited to Hraness; every claim links to its cataloged source.