Evidence-first authority page
What Is an Open Source AI Agent? Definition and What to Check
A definition of open source AI agents, what 'open source' actually covers in an agent stack, and what to verify before adopting one.
Direct answer
An open source AI agent is one whose orchestration code — the logic that decides how the agent reasons, calls tools, and manages state — is published under an open source license, letting anyone inspect, modify, self-host, and redistribute it. Open source agent code does not automatically mean the underlying model it calls is also open source.
Last reviewed 2026-08-29
This page remains noindex while its page-level source receipts are completed and verified. It is available for review, not presented as publication-ready authority.
Three layers to check separately
- Agent orchestration code — the framework or application logic (e.g. Cline, Aider). Check the actual license file, not just a marketing claim of 'open source.'
- The underlying model — many open source agents call a closed, proprietary model via API by default, even though the orchestration layer is fully open; using an open source agent does not imply an open source model unless you specifically configure a local or open-weight model.
- Any hosted service layer — some 'open source' agents also offer a hosted, paid version with additional closed-source features on top of the open core.
Why the distinction matters
Teams choosing 'open source' for auditability, self-hosting, or vendor-independence reasons need all three layers to match that goal — an open orchestration layer calling a closed, rate-limited third-party model API still leaves you dependent on that API vendor for the part of the stack that matters most for availability and cost.
Conversely, an agent that's technically open source but has an inactive repository, no recent commits, or a maintainer who has moved on carries a different kind of risk — access to the source doesn't guarantee it stays maintained.
What to verify before adopting one
- The actual license (MIT, Apache 2.0, and copyleft licenses like AGPL carry different obligations for redistribution and hosted use).
- Which model(s) it calls by default, and whether local/open-weight models are a supported, first-class option rather than an afterthought.
- Recent commit activity and issue-response time as a proxy for active maintenance.