Documentation

Search.co is a multi-agent research OS: BYOK model routing, reusable research DAGs, and GitHub-native output.

BYOK providers

Add keys for OpenAI, Anthropic, Gemini, Groq, DeepSeek, Mistral, xAI, Z.ai, OpenRouter, local endpoints, and more. Keys are encrypted at rest (AES-256-GCM) and never displayed after saving. With no key, runs use a deterministic mock provider — free and offline.

Cost-optimizing router

Every agent task is routed to the cheapest CAPABLE model given its reasoning level, context needs, output format, and sensitivity — with automatic fallback to another provider if one fails, recorded in the run logs.

Research DAGs

A workflow is a DAG of agent nodes. Nodes run in parallel where dependencies allow. Start from templates (Company, M&A, SEO, Open-Source Repo, Legal, Market Landscape) or compose your own.

Source connectors

Modular connectors — web search (Tavily/Exa/Brave/SerpAPI), SEC EDGAR, GitHub, arXiv, PubMed, Reddit, sitemap, RSS, custom URL — all behind one interface with SSRF-guarded fetching.

GitHub-native output

Each run produces an executive summary, findings, sources, entities, risks, cost breakdown, a full Markdown report, and machine-readable JSON — committed to your repo (direct, PR, or preview) under a versioned /research tree.

Getting started

Sign up, add a provider key (optional), connect a GitHub repo (optional), then run Company Research on a target like “NVIDIA”. Watch the DAG execute, the router pick models, and the report land in your repo.

Full setup + architecture notes live in the repository README. Ready? Create an account →