Open research agent with multi-provider routing, live tool use, and full citations.
Ask a question → the agent searches, reads sources, reasons, and returns a cited answer in real time.

Website · X · GitHub

GrayTell is a research agent, not a chatbot.

When you ask a question it:

1. Decides which tools to use
2. Executes them in real time (you see every step)
3. Reasons over the evidence
4. Returns a structured answer with citations

It routes across 5 providers and 30+ models through a single interface, with a fallback chain so tools keep working even if a premium search provider is unavailable.

• Agentic tool loop — up to 8 rounds (Search mode) or 12 rounds (Deep mode), with parallel tool calls
• 12 research tools — web search, page reader, site map, Reddit, X, GitHub, YouTube, LinkedIn, and more
• Multi-provider routing — Groq, OpenRouter, Mistral, Ollama Cloud, ZAI
• Live research timeline — watch the agent think, search, and read as it works
• Streaming NDJSON protocol — full visibility into every step
• Citations by default — every claim can be traced to a source

graph TB
    subgraph Client ["Browser"]
        UI[Chat UI]
        Timeline[Research Timeline]
    end

    subgraph Backend ["Next.js App Router"]
        ChatAPI["/api/chat"]
        AgentLoop[Agent Loop]
        ProviderRouter[Provider Router]
        ToolRuntime[Tool Runtime]
    end

    subgraph Providers ["AI Providers"]
        Groq[Groq]
        OpenRouter[OpenRouter]
        Mistral[Mistral]
        Ollama[Ollama Cloud]
        ZAI[ZAI]
    end

    subgraph Tools ["Research Tools"]
        WebSearch[Web Search]
        PageReader[Page Reader]
        SocialSearch["X / Reddit / GitHub / ..."]
    end

    UI --> ChatAPI
    ChatAPI --> AgentLoop
    AgentLoop --> ProviderRouter
    AgentLoop --> ToolRuntime
    ProviderRouter --> Groq
    ProviderRouter --> OpenRouter
    ProviderRouter --> Mistral
    ProviderRouter --> Ollama
    ProviderRouter --> ZAI
    ToolRuntime --> WebSearch
    ToolRuntime --> PageReader
    ToolRuntime --> SocialSearch

• Search mode: max 8 rounds
• Deep mode: max 12 rounds
• Multiple tool calls can run in parallel each round
• If the round limit is hit, the system forces a final answer

{"type": "meta", "model": "mistral-medium-latest", "mode": "deep"}
{"type": "step.start", "stepId": "s1", "tool": "exa_search", "label": "Web search"}
{"type": "step.thinking", "stepId": "s1", "text": "I need recent information about..."}
{"type": "step.input", "stepId": "s1", "queries": ["..."]}
{"type": "step.output", "stepId": "s1", "results": [...]}
{"type": "step.done", "stepId": "s1"}
{"type": "answer.delta", "text": "Based on the sources..."}
{"type": "answer.done"}

GrayTell does not lock you to one provider.

Supported providers:

• Groq — ultra-fast inference
• OpenRouter — broad model marketplace
• Mistral — strong reasoning models
• Ollama Cloud — open-weight models
• ZAI — GLM family + built-in tools

Custom adapter converts Ollama’s native NDJSON format into OpenAI-compatible SSE so the rest of the system stays uniform.

• Frontend: React + Next.js App Router
• State: Zustand
• Backend: Next.js API routes + streaming NDJSON
• Auth / DB: Supabase + Prisma
• Tools: Exa (optional) + ZAI fallback chain

Detailed setup instructions coming soon.
For now, see the website: graytellai.space-z.ai

GrayTell is early-stage and under active development.

Current focus:

• Improving citation faithfulness
• Better tool selection and error handling
• Clearer research timeline
• Open evaluation of agent reliability

We are not claiming to be a full AI safety lab yet. We are building the foundations for more transparent and verifiable research agents.

• Website: graytellai.space-z.ai
• X: @GrayTell_Org
• GitHub: github.com/GrayTell

Apache License 2.0

Copy everything above and replace the entire contents of your `README.md` with it.