Sweep AI is an open-source GitHub-native coding agent (MIT, YC S23) that converts labeled GitHub issues into pull requests automatically. Triggered by adding a "sweep" label to any issue, it searches the codebase using lexical + vector retrieval, plans multi-file changes, generates code, runs linters and tests, and submits a PR for human review. 7.6K GitHub stars, ~$1M ARR with a five-person team, and 40K+ JetBrains plugin installs indicate real traction for a focused niche.
Note: There is a name collision — a separate company also called "Sweep" (founded 2021) focuses on Salesforce/HubSpot metadata and raised a $22.5M Series B in May 2025. This entry is for the AI coding agent at sweep.dev, not that company.
Why It's in Assess
Sweep has been operating since 2023 and has genuine adoption — but it occupies a narrower, less ambitious position than most coding agents on this radar. Its strength is handling well-defined, discrete GitHub issues (bug fixes, config changes, small features) without developer supervision. Its limitations define its ceiling.
Assess rather than Trial because:
- GitHub-only by design: GitLab support exists but is community-maintained and less mature. No native Linear, Jira, or Azure DevOps integration as the primary trigger surface (Jira support added as secondary)
- No published SWE-bench scores: Unlike Devin, OpenHands, or Open SWE, Sweep has not published benchmark results on the standard SWE-bench leaderboard
- Focused scope ceiling: Performs well on discrete, well-described tickets; struggles with ambiguous requirements or tasks requiring deep architectural judgment — the same ceiling as its 2023 design
- Limited enterprise case studies: YC profile lists Brex, NBC Sports, LG Electronics, Mass General Brigham — but no detailed production metrics have been published
Sweep's sweet spot is teams using GitHub Issues as their task surface who want lightweight issue-to-PR automation without the infrastructure overhead of Symphony or the enterprise cost of Devin. For this use case, it is usable today.
When Sweep is the right choice: GitHub-native teams wanting simple, zero-infrastructure issue-to-PR automation on well-defined tickets. Open-source projects where contributors want to assign mechanical tasks to an agent.
When to look elsewhere: Complex, multi-step architectural work (use Devin or OpenHands); Linear-first teams (use OpenAI Symphony); teams needing enterprise SLAs (use Factory AI); teams wanting the highest SWE-bench performance (Claude Code, OpenHands).
How It Works
- Add the
sweeplabel to any GitHub issue (or @-mention Sweep in a comment) - Sweep reads the issue description and searches the codebase via lexical + semantic vector retrieval
- A planning phase produces a multi-file change plan
- Code is generated and validated against linters, autoformatters, and GitHub Actions CI
- A PR is submitted with the changes for human review and merge
- Optionally available as a JetBrains plugin for inline autocomplete
Key Characteristics
| Property | Value |
|---|---|
| Interface | GitHub App (webhook), JetBrains plugin, CLI (PyPI) |
| Provider | Sweep AI (YC S23) |
| License | MIT (open-source core); Enterprise Edition available |
| Pricing | Free tier; Plus ~$120/mo (30 GPT-4 tickets); Team: $40 Sweep API credits/seat; Enterprise: custom |
| Underlying model | GPT-4 (default); configurable |
| Sandbox | E2B-compatible execution for validation |
| SWE-bench | Not published |
| GitHub | sweepai/sweep |
| Website | sweep.dev |
| Docs | docs.sweep.dev |