01 / Build
Move a feature forward.
Describe a change or start from a connected issue. Approve the plan, then follow the implementation and review its results.
Example request
Add CSV export to the reports page.
Self-hosted autonomous development
AI coding that can’t fake done.
Build features, fix errors, and improve your codebase on your infrastructure. CrewWork executes the work and runs independent validation. You review the changes and the evidence before deciding what merges.
Local open-weight models. Required checks. Human review.
Work, decisions, and evidence
View
Build. Repair. Improve.
One workspace for the changes you want to make and the problems you need to solve.
Explore autonomous coding
02 / Repair
Act on the errors you find.
Turn runtime errors and security findings into proposed fixes. Inspect the changes and required validation before deciding what merges.
Example request
Fix the checkout error reported by Sentry.
03 / Improve
Give maintenance a next step.
Start from a measured coverage gap or a code improvement suggestion. Scope the work to a specific target and inspect the outcome.
Example request
Add tests for the uncovered paths in billing.
Delegate the work. Stay close to the decisions.
A clear path from a connected repository to a change you can review.
- 01
Connect your code
Connect a repository or start a local project. CrewWork indexes the code to give each run project context.
- 02
Approve the plan
Describe the work or start from a finding. Review the proposed scope and validation targets before execution.
- 03
Follow the run
CrewWork implements the change and runs required checks, with repair attempts bounded by your configured limits.
- 04
Review the evidence
Inspect the changes, executed checks, and results. You decide what merges into your project.
An inbox for your attention.
- Needs you
- Find decisions, failures, and work that needs your attention.
- Active
- Follow runs and background jobs already in progress.
- Done
- Return to completed work and inspect its recorded outcome.
Saved run state and checkpoints support recovery after interruptions. Time, token, and iteration budgets bound execution, while workspace locks coordinate repository writes.
Walk through the full workflowIndependent verification
Evidence behind every completion.
- Checks run independently
- Completion depends on validation evidence produced by CrewWork. A model’s own report of success cannot satisfy that requirement.
- Missing evidence stops completion
- Required checks must run and pass. Failing or missing evidence keeps a run from being marked complete on the model’s say-so.
- You can inspect the result
- See the changes alongside executed checks, logs, and acceptance results. Decide what to merge after reviewing the evidence.
Passing checks establish what was verified. They do not guarantee that every possible defect has been found.
Read how verification works
Changes and validation evidence
View
1/2
The same scrutiny for its own code.
Platform Self-Repair uses Qwen3.8 27B to propose changes to CrewWork itself. Isolated validation and a separate promotion test matrix check the candidate before a draft pull request reaches a person for review.
Explore Platform Self-RepairLocal by default
Your infrastructure. Your model choices.
CrewWork uses the same Qwen3.8 27B open-weight model for planning, coding, review, structured extraction, and chat. No frontier API is required. You can configure other OpenAI-compatible endpoints, with public model endpoints requiring an explicit opt-in.
What self-hosting involves- Application
- One Docker Compose host runs the workspace, workers, and data services.
- Generation
- The same 27B model can serve different workloads through separately configured model endpoints.
- Code search
- A separate embedding model supports semantic search. Model-serving hardware is configured separately from the application host.
Bring your next piece of work.
Discuss what you want to build, repair, or improve, along with your deployment needs and access to CrewWork.
Or view the architecture for the full technical picture.