The Graytell research agent
An open research agent with multi-provider routing across 30+ models, live tool use in up to 12 rounds, and citations by default.
- Date
- 2025
- Category
- Open source
- Details
- Agent
AI will have a vast impact on the world. Graytell is an AI safety research lab dedicated to securing its benefits and mitigating its risks — publishing research you can cite and shipping products you can run, all in the open.
Featured product
The frontier monochrome design system for everything you code — 168 copy-paste React components: 55 shadcn/ui primitives, all 48 AI elements and 12 agent blocks. Rides the standard shadcn CLI.
An open research agent with multi-provider routing across 30+ models, live tool use in up to 12 rounds, and citations by default.
An empirical evaluation of citation accuracy in DeepSeek, Claude and ChatGPT under research-agent conditions — 120 claims, raw responses and scoring data published.
The lab's first product — dictation-first deep research with 30+ models across 5 providers, real-time tool calling, and streaming answers.
Graytell Labs · AI safety research
Graytell Labs is an AI safety research platform. We publish research you can cite — and ship working products you can run today. Every one of them is open source.
The open research agent. Multi-provider routing, live tool use, and full citations — ask a question, watch every step, check every source.
Apache-2.0 · TypeScript
The canvas for your next interface — 168 copy-paste React components in a strict monochrome design system. 55 shadcn/ui primitives, all 48 AI elements, 12 agent blocks.
MIT · shadcn-style registry
Where Graytell began: a dictation-first deep-research agent with 30+ models across 5 providers, real-time tool calling, and streaming answers.
Live · the lab's first product
Everything Graytell publishes is open source — read the paper, run the research agent, or build your next interface with the 168-component Gray UI registry.