Features
Build features, repair errors, and improve code quality in one workspace. Define the work, follow execution, and inspect changes with independent validation evidence. You decide what merges.
Plan and Deliver
From a requested feature to changes you can review
Describe the outcome you want or start from a connected issue. CrewWork proposes a plan for your approval, implements the change, and runs required validation. Repair attempts happen within the run, followed by a final review of the whole program. See how it works.
- Plan approval: review the proposed scope, risks, assumptions, and validation targets before execution.
- Greenfield-ready: seed a starter (FastAPI, Flask, Express, React, Vue, Svelte, or Astro) or start from an empty repo.
- Configurable limits for time, iterations, and tokens, with automatic detection of when work is done.
- Every run is inspectable and resumable, its attempt ledger showing validation, repair, and review events inline.
- Automated test validation across 10 stacks: Python, JS/TS, Go, Rust, Ruby, PHP, .NET, JVM via Maven/Gradle, C/C++ via CMake/CTest.
Runtime Error Repair
From runtime error to a validated, reviewable fix, automatically
Error monitoring connects runtime errors to proposed fixes and their required validation. Frequent errors become repair candidates through an opt-in trigger with cooldown windows and daily budgets. Fixes land on isolated local delivery branches; push is disabled by default.
- Repair retries resume the prior attempt's memory instead of cold-restarting.
- A STAGED status marks a fix verified on a branch but not yet published.
- A successful fix automatically resolves the linked error-monitoring issue.
- Full logs and diffs attached for every fix, plus repair scheduling from preview failures.
- Failed attempts stay isolated from the target branch, with run status and diagnostics available for inspection.
Source Control & Preview
Review AI-generated changes and watch them run, without switching tools
A selected Project’s contextual Code workspace shows commit history, branch context, and per-file diff review. Preview is a top-level, project-selected destination that runs the application and feeds runtime signals into repair. Its quality evidence measures Core Web Vitals, accessibility, and bundle size, and every preview container gets a signed identity with egress blocking. See source control & integrations.
- Project-scoped git graph and commit timeline with inline diff review.
- Compare view: diff any earlier commit against the selected revision.
- Start, restart, and stop preview containers with live runtime verification.
- Preview quality gate: Core Web Vitals, accessibility, bundle size, and console errors on the exact revision.
- Signed, network-isolated container identity with active egress blocking to protected infrastructure.
Delivery Runtime
Reliable multi-step execution with budgets, checkpoints, and automatic completion detection
CrewWork tracks execution, budgets, and Git checkpoints so an outcome recovers from saved state. Pause or resume supported work, inspect its evidence, and see why it stopped. See the architecture.
- Run state and project-level progress logs survive worker restarts.
- Start work globally, from CrewMate, or from a Project and its grounded evidence.
- Repeated identical failures stop after three consecutive occurrences, with the count preserved across restarts.
- Full task summaries and logs for every run.
Continuous Improvement
Turn coverage gaps and code findings into scoped work
CrewWork identifies coverage gaps and code improvement opportunities, then proposes scoped work. Suggestions are reviewed for impact, risk, and confidence before reaching your queue. Start a coverage improvement run from a measured file and revision in Quality.
- Generates single-PR-sized recommendation tasks with impacted files and validation footprint.
- Evidence required: each suggestion cites real file paths and code excerpts.
- An Improve action starts a reviewed run scoped to a low-coverage file at its measured baseline.
- Tracks applied vs. dismissed outcomes for feedback loops.
Reliability and Governance
Automation that survives restarts, respects budgets, and keeps every decision auditable
Workspace locks prevent conflicting automation, and every remediation writes an append-only event log auditors can query later. Validated builds can deploy to hosts on your own network through a mutual-TLS release agent, with on-call escalation if a deployment fails. See releases & on-call.
- Workspace locks coordinate repository writes and expose lock metadata in the API.
- LAN release agent: outbound-only, mutual TLS, no inbound connectivity to the target.
- Ingested error events are retained seven days by default and can be replayed with a dry-run preview.
- Compliance evidence reports: non-certifying, project-scoped views across eight evidence areas.
The Workbench
A full development workbench anywhere you have a browser. Agents write the code; you review it
The persistent rail contains Work, Projects, Security, and Preview. A selected Project opens contextual Code with Files, Source Control, Intelligence, and a read-only Monaco code reader. Project settings, Organization, Account, and permission-gated CrewWork Admin are utilities rather than destinations. The CrewMate assistant panel answers questions grounded in your repository and speaks MCP directly to external tools. See platform & safety.
- File explorer with directory navigation and multi-tab file review.
- Command palette with quick open, symbol search, and repository clone.
- Spec-compliant MCP server: external clients read files, search code, and inspect git status, read-only.
- Charter-driven programs and a suggestion pipeline generate persistent task queues.
Work, Outcomes, and Evidence
See what needs attention, what CrewWork is doing, and what result it produced
Work opens on Needs attention and an Outcomes collection. Start work creates one planned outcome or direct request; opening it shows the narrative, changes, logs, diagnostics, validation evidence, and review decision. See how it works.
- Open an outcome to its execution narrative, changes, validation evidence, diagnostics, and logs.
- Typed runs, jobs, action items, and campaigns stay directly addressable without becoming Work sections.
- Filter outcomes by project, or follow the whole collection from one place.
Code Intelligence
Search and navigate your codebase by meaning
Search your codebase by intent, for example "where do we handle user authentication," and get relevant results. Structural analysis builds a symbol graph of functions, classes, and methods, with their imports, calls, inheritance, and complexity metrics, so search and planning follow real dependencies, not text similarity alone. Jump to definitions, find references, and trace relationships, and ask CrewMate to explain undocumented symbols. See the architecture.
- Go-to-definition and find-all-references across your entire codebase.
- Multi-language support: Python, JavaScript, TypeScript/TSX, Go, Rust, Java, C/C++.
- Duplicate-code detection surfaced in your project Health Score.
- Reviewed refactoring recipes with a hash-verified preview before any change applies.
- Cross-project semantic search and batch refactoring across projects you can access.
Security Scans
Know your vulnerabilities before attackers do
CrewWork runs project security scans across 18 scanner profiles: SAST for Python, JS/TS, Go, Ruby, C/C++, and Java, dependency and license checks, secret scanning, infrastructure-as-code analysis, and opt-in DAST (passive baseline, plus bounded active web and API modes) against your running preview. See security.
- Reachability evidence attached to dependency findings.
- SARIF, SBOM (SPDX), and OpenVEX export, plus SARIF import as the canonical latest scan.
- A findings triage queue with a per-finding inspector showing fix eligibility and the active remediation attempt.
Provider Integrations
Fits into the git workflow your team already uses
CrewWork connects to GitHub, GitLab.com, Bitbucket Cloud, and Jira Cloud, one connection per provider per organization. GitHub gets the deepest automation, including pull request authoring. Project fixes push only when enabled.
- OAuth or API-token connections, with sign-in identity kept separate from organization connections.
- Automatic PR review comments with ownership, CODEOWNERS gaps, and blast-radius impact analysis.
- The five-part quality gate posts as its own GitHub status check on the same commit.
See whether CrewWork fits your work.
CrewWork is built and running, and not yet generally available.
- worth discussing
- The errors and backlog you would hand off
- Where your source has to stay
- The evidence you would need to trust a change