argo

If you've built agents, you know the pain: built-in model search gives you a wall of text, AI search gives you a pretty summary page, but your agent actually needs ranked, verifiable, context-friendly evidence. Argo treats search as an evidence pipeline — language detection, domain routing, multi-engine recall, RRF fusion, quick evidence scoring — and delivers compact JSON with credibility breakdowns, not link lists. Agents can sort it, fetch it for verification, and fit it in context without blowing up.

Argo's superpower is “ask what it looks like”: 150+ sources across 70+ domains. Stock quotes hit snapshot sources directly, movies route to IMDb, chemical formulas go to PubChem, US CPI queries flow to macro data feeds; Japanese and Korean searches stay clear of Chinese-only engines. It doesn't just find things — it verifies: evidence scores, fetch_required gates for high-stakes questions, and --verify to backfill post-check scores. Deep research breaks broad questions into sub-questions, collects from multiple sources in parallel, and tells you what evidence is still missing. Dual-layer caching returns hot queries in ~10ms, and 12 MCP tools cover search, fetch, screenshots, PDFs, social listening, local file search, and even a sandboxed recompute executor.

Who is it for? Anyone running agents who is tired of “search quality is a dice roll and context gets flooded with text.” Whether you use Claude Code, Cursor, Cline, or your own Python project, Argo plugs in via MCP or as a library. Zero-config to start — free engines first, API keys all optional, four budget modes from fast daily lookups to full research depth. It doesn't replace your toolchain; it just fixes what search results should look like — turning every agent retrieval into evidence, not noise.