Morph
@morphllm
Followers
2K
Following
847
Media
30
Statuses
284
ultra fast models that improve coding agents 10,500 tok/sec. https://t.co/dBQjovGya3
San Francisco, CA
Joined May 2025
Introducing WarpGrep, a fast context subagent that improves coding agent performance. WarpGrep speeds up coding tasks 40% and reduces context rot by 70% on long horizon tasks by treating context retrieval as its own RL trained system. Inspired by Cognition’s SWE-Grep - we’re
62
84
1K
we just shipped an update to WarpGrep that includes some breaking changes. to update, run npm cache clean --force && rm -rf ~/.npm/_npx and reload your coding agent
3
0
18
v1 soon!
We’re releasing v1 of WarpGrep soon! This will be production quality and ready for teams to integrate. Our latest evals on this version put WarpGrep v1 ahead of Claude Haiku and SWE-Grep
0
0
15
Hey @morphllm meet @FactoryAI , Factory meet Morph... I have Droids running with the new Morph MCP and ... it's glorious -- Details below on some edits I made to https://t.co/uYPguFyjPB and https://t.co/HHhllTEqke to make it stick better
9
11
115
gonna be posting some indepth evals super soon, but here's a teaser: - Stock Claude Code spent a minute on search, with warp grep it spent 14 seconds - Stock Claude Code injested 75k tokens, with warp grep it injested 3k - They both returned identical results (down to the line
Introducing WarpGrep, a fast context subagent that improves coding agent performance. WarpGrep speeds up coding tasks 40% and reduces context rot by 70% on long horizon tasks by treating context retrieval as its own RL trained system. Inspired by Cognition’s SWE-Grep - we’re
15
11
284
I’ll be going over how we got sub 2s inference on ~90k code input (served on multiple b200s) in the next few days!
Introducing WarpGrep, a fast context subagent that improves coding agent performance. WarpGrep speeds up coding tasks 40% and reduces context rot by 70% on long horizon tasks by treating context retrieval as its own RL trained system. Inspired by Cognition’s SWE-Grep - we’re
9
3
161
Super hyped to ship this. RL is notoriously compute inefficient (even more so with MoEs), here's a thread on how we got around it
Introducing WarpGrep, a fast context subagent that improves coding agent performance. WarpGrep speeds up coding tasks 40% and reduces context rot by 70% on long horizon tasks by treating context retrieval as its own RL trained system. Inspired by Cognition’s SWE-Grep - we’re
2
3
43
we'll be shipping updates often as training progresses! v1 coming soon!
1
0
13
You can install WarpGrep into Claude Code, Codex, or your favorite coding agent today! Use code BF16 for 40 million tokens of credit
morphllm.com
One MCP. Plug into Cursor, Claude Code, or any agent. Faster edits, smarter retrieval, and better context.
5
3
96
SWE-Grep runs at around 650 tokens per second on Cerebras. WarpGrep hits around 900 tokens per second on B200. We worked closely with NVIDIA to optimize WarpGrep. CUDA gives us the stability and customization ability to do weird things stably.
1
0
51
We’re bullish on subagents that do parallel tool calls and making task specific inference engines for them. For WarpGrep: - a strict budget of 4 turns - up to 8 parallel tool calls per turn (grep, list, read, etc.) - Inference optimized for grep (super prefill heavy) a reward
1
0
35
We found that giving models access to warp grep boosts performance between 5-12% with frontier models, while improving speeds and reducing token usage up to 40% WarpGrep is our attempt to fix coding agents.@swyx hit the nail on the head with the semi async valley of death.
2
1
43
Context pollution does more than just waste tokens, it actively kills model performance. In production repos it poisons the model’s thinking into editing files you didn’t want to touch, while costing exponentially more. The WarpGrep approach: have a inference optimized model
1
1
49
Coding Agents spend 60% of their time searching instead of coding. We built WarpGrep so people can have sub 10 second agents, without compromising on accuracy. Dev flow-state is important. P(bounce) increases 10% with each second a dev has to wait according to @swyx and
1
6
96
Create your own "starter" version of Lovable in under 6 minutes. Here's how 👇 You'll need API keys/accounts with: - @firecrawl_dev - @GeminiApp - @vercel - @morphllm p.s. keys have been rotated :)
Introducing Open Lovable v3 - an open source AI design engineer 🔥 Enter a URL and agents will create an on brand clone you can build on top of. Powered by Gemini 3 Pro and @firecrawl_dev's new branding format. Try out the 100% open source example, link below!
3
3
22
The coding agent in Pilot generates the feature flag code along with the UI updates and uses @morphllm to apply the edits to your files. Super-quick and accurate.
1
1
8
In 2026, coding agents will go mainstream to the masses. This won’t mean everyone will be vibecoding. Sufficiently fast and good codegen is the new modality of creative expression. State, reactivity, and interactivity will make disposable codegen miniapps a pillar of online
0
0
8
