Tickets in. Shipped code out.
Three AI agents refine your specs, write your code, and review your PRs — right inside Linear and GitLab.
Your tickets aren't ready. Refine makes them ready.
Refine reads your codebase, analyzes screenshots and Loom videos, then asks targeted questions until the issue is airtight. You get acceptance criteria, risk analysis, and calibrated story points — not guesses.
- Codebase-aware Q&A
- Loom & screenshot analysis
- Acceptance criteria + story points
Refine
What error handling should apply?
You
Use retry with exponential backoff
Refine
✓ Added to acceptance criteria
Refined ticket in. Merge request out.
Build reads the refined spec, understands your codebase conventions, and writes the implementation. It creates a branch, commits clean code, runs tests and lint, and opens an MR — then tells you exactly how confident it is.
- Writes code & opens MRs
- Runs tests, lint & build
- Confidence score (0-100%)
Every MR reviewed. Every line scored.
Review inspects merge requests with a 6-category scoring rubric — correctness, product behavior, security, quality, tests, and completeness. It posts inline comments, checks your CI pipeline, and approves or blocks with a confidence-backed risk assessment.
- 6-category weighted scoring
- Inline comments with severity
- Approve or block with confidence
Overall
87/100
One ticket. Three agents. Shipped code.
Ironwork runs the full pipeline — from vague ticket to merged, reviewed code — without your team context-switching.
@mention them like a teammate.
Ironwork agents are native Linear agents. Assign Refine to an issue. Trigger Build from a comment. No dashboards, no CLI, no context switching.
Assignees
Your code. Your keys. Your control.
Encrypted Isolation
Every workspace is fully isolated. Repos encrypted at rest. Agents only access code during active tasks.
Bring Your Own Key
Use your Anthropic API key. No markup, no middleman. Full cost visibility on your Anthropic dashboard.
Multi-Tenant Architecture
Workspaces, roles, SSO-ready auth. Built for teams, not solo experiments.
Free. Seriously.
Ironwork is free to use. You bring your own Anthropic API key and pay Anthropic directly. No platform fees, no per-seat pricing, no surprises.
Free to use. You control the costs.
Ironwork is free. You bring your own Anthropic API key and pay Anthropic directly for Claude usage. No markup, no hidden costs.
- No platform fees
- No per-seat pricing
- Full cost transparency