Loading the catalogue…
Coding agents · Princeton NLP (Princeton University / Stanford University researchers) · US
SWE-Agent
US CLOUD Act exposureOpen sourceMIT
STAV REVIEW · LAST REVIEWED 06/21/2026
“Fully self-hosted with no vendor infra in the path, but an active deprecation notice and unrestricted shell access make this a project to evaluate carefully before committing.”
Good for
- Research and security teams who need a battle-tested, MIT-licensed autonomous coding agent they can run entirely on their own infrastructure
- Organisations that want direct control over Inference routing — the LLM provider is fully operator-configurable, so data boundaries are set by your own choice of endpoint
- Teams benchmarking autonomous coding agents against academic baselines, given the NeurIPS 2024 publication and 19,500-plus GitHub stars
- Proof-of-concept deployments where the deprecation posture is acceptable and a migration path to the successor project is already planned
Watch out for
- The agent runs with unrestricted bash and vim access on the host; sandboxed container execution with explicitly scoped filesystem and network permissions is mandatory — direct host execution is not safe for enterprise use
- The vendor's documentation carries a live, site-wide banner recommending migration to mini-swe-agent; no release has shipped since v1.1.0 in May 2025, and Docker Hub images have not been updated since December 2024 — any adoption must include a documented migration plan
- All prompt and code content exits the enterprise perimeter to your chosen LLM provider API; data residency and retention are governed entirely by that provider's own terms, not by SWE-agent — use the Smart Router to enforce only DPA-covered Inference endpoints
Editorial assessment, not legal advice. Stav's risk ratings, scores, and verdicts are our own analysis of publicly available information and may be incomplete or out of date. Verify independently before making compliance or procurement decisions.
MODEL REQUIREMENTS
What SWE-Agent needs from a model.
Required
- Tool calling
Recommended
- Code-native
- Long context (≥128k)
Known issues2 open
Public security, privacy and governance advisories that are still open. Each row carries a source link — these are factual claims, not editorial judgements.
Data flow & telemetry
How SWE-Agent moves data between the client, the inference endpoint and any vendor-owned services.
Data flow
- Direction
- local_only
- Stores prompts
- No
Telemetry
- Collects telemetry
- false
Vendor & licensing
- Developer
- Princeton NLP (Princeton University / Stanford University researchers)
- Vendor country
- US
- Vendor type
- foundation
- Funding status
- foundation_backed
- Runtime type
- cli
- Open source
- Yes
- Licence
- MIT
- Licence risk
- permissive
- CLOUD Act exposure