AI agent rankings — by the one metric everyone publishes.
Open-source agent projects ranked by GitHub stars: an adoption and attention signal, snapshotted with its date from the sources cited on each profile. Closed products are listed separately, unranked — on purpose.
| # | Project | Type | GitHub stars (as of) | License | Status | Repository |
|---|---|---|---|---|---|---|
| 1 | OpenClaw | Product | 388,000 (2026-09) | MIT | Active | openclaw/openclaw |
| 2 | Hermes Agent | Product | 249,000 (2026-09) | MIT | Active | — |
| 3 | OpenAI Codex | Product | 62,000 (2026-10) | Apache-2.0 (CLI) | Active | — |
| 4 | AutoGen | Framework | 61,243 (2026-09) | MIT (code); documentation CC-BY-4.0 | Maintenance mode | microsoft/autogen |
| 5 | CrewAI | Framework | 59,233 (2026-09) | MIT | Active | crewAIInc/crewAI |
| 6 | LlamaIndex | Framework | 52,300 (2026-09) | MIT | Active | — |
| 7 | LangGraph | Framework | 42,526 (2026-09) | MIT | Active | langchain-ai/langgraph |
| 8 | OpenAI Agents SDK | Framework | 29,789 (2026-09) | MIT | Active | openai/openai-agents-python |
| 9 | Microsoft Agent Framework | Framework | 14,000 (2026-10) | MIT | Active | microsoft/agent-framework |
| 10 | Strands Agents | Framework | 8,300 (2026-09) | Apache-2.0 | Active | strands-agents/harness-sdk |
Read this table correctly: stars accumulate from launch day and are never removed, so they reward age and hype as well as present-day use. AutoGen ranks near the top on stars while in maintenance mode — its status column says so. Downloads, production usage and revenue would rank these projects differently, but vendors do not publish them on a comparable basis, so we do not fake a composite. Star figures are dated snapshots from the sources on each project's profile page, checked 2026-10-14.
Not ranked: closed agent products
These products publish no adoption metric that can be compared honestly — no public repository, no shared benchmark, no audited user counts. Listing them by category with their pricing model is the truthful treatment; anyone selling you a single ranking of these products is estimating in private.
| Product | Category | Pricing model |
|---|---|---|
| Devin | Coding agents | Subscription + usage credits |
| Manus | General-purpose agents | Subscription + credits |
| Claude Code | Coding agents | Bundled subscription / API usage |
| Cursor | Coding agents | Subscription + usage credits |
| GitHub Copilot | Coding agents | Per-seat subscription |
Building on a model instead? The model database and the League standings rank the models these agents run on.