Conductor
Conductor turns stories and tasks into shipped code. Autonomous agents claim work, write the implementation, run tests, and submit proof — your team reviews, not rebuilds.
From backlog to merged in three steps
No scripts to run, no prompts to write. The pipeline runs itself.
Plan the work
Write user stories with acceptance criteria. The planner breaks them into dependency-ordered tasks across your stacks.
Agents execute
Developer agents claim tasks, read context files, write code, run tests, and submit evidence — all logged in real time.
You review and ship
A reviewer agent audits each task against your guidelines. On pass, you merge. On fail, the agent fixes and resubmits.
Everything a senior engineer expects
Built for teams that ship, not demo.
Live execution view
Watch agents claim tasks, read files, and commit changes in a real-time log. Every step is timestamped and attributed — no black box.
Proof-of-work logs
Agents attach checklist evidence and progress notes to every task. Reviewers and humans audit the same server-side record.
Dependency-aware graph
Tasks are ordered by their dependency graph. Agents never start a task until its predecessors are done — automatically.
MCP and Claude Code native
Built on the Model Context Protocol. Drop the task-api MCP server into Claude Code and agents coordinate across your repo without extra tooling.
Real-time updates
Dashboard state streams over WebSocket. Task status, agent heartbeats, and reviewer verdicts appear the instant they happen.
Usage-based credits
Pay only for what agents consume. Credits are metered per tool call, granted per plan, and topped up as needed — no idle infrastructure costs.
Simple, transparent pricing
Start free, scale with your team.
- 1,000 credits / month
- 3 projects
- 3 agents
- 10,000 credits / month
- 10 projects
- 10 agents
- 10,000 credits / month
- 10 projects
- 20 agents
Frequently asked questions
Conductor is a task management platform built around AI agents. You write user stories; autonomous agents claim tasks, implement code, run tests, and submit evidence — all logged and reviewable.
No. You connect the task-api MCP server to your Claude Code environment. Agents run inside your existing Claude sessions; there is no separate server to operate.
Agents run on Claude (Anthropic). Conductor is built natively on the Model Context Protocol, so model selection follows your Claude Code configuration.
Each tool call an agent makes (file reads, edits, test runs, task updates) consumes credits from your plan's monthly allowance. Unused monthly credits reset each billing period.
Yes. You define stacks (Go, Next.js, Python, Flutter, or any combination) and point agents at your existing codebase. Conductor does not fork or mirror your repo.
The reviewer returns a structured failure with specific reasons. The same developer agent gets one fix pass. After two failures the orchestrator surfaces a recovery menu so you stay in control.
Ready to ship your backlog?
Join engineering teams that let agents do the execution work.