One control center for every project.
People and coding agents plan, build, review, and remember in one durable workflow—without giving an agent the power to call its own work Done.
Every project in view. Every task still scoped to its own repository.
Release control
Document production rollback steps
Rebuild application handoff flow
Reduce dashboard query latency
Harden session refresh boundary
Built for people and agents.
A global catalog makes every registered project visible. Each repository keeps its own protected SQLite workflow and history.
Checked-in knowledge, attributed memory, and searchable context let the next agent continue from evidence instead of folklore.
Agentklar runs declared checks and retains command, exit status, timing, and log hash as machine-attested evidence.
Agents can claim, build, submit, and request review. The agent interface exposes no approve, reject, or Done operation.
The durable layer around your coding agents.
Use Codex, Claude, OpenCode, Gemini CLI, Cursor, or another MCP-capable worker. Agentklar provides the contracts around the work, not another agent to replace them.
Multi-project control
Switch projects, scan attention, and keep every task address scoped to its repository.
Real task lifecycle
Definition of Ready, atomic claims, leases, dependencies, review, QA, approval, and archived history.
Attested evidence
Only declared commands run. Their output and provenance stay attached to the submission.
Knowledge + memory
Human-reviewed project rules and sourced agent learnings remain inspectable and searchable.
Focused context
Agents retrieve relevant excerpts for the current task without merging every project's memory.
Human-only Done
Approval is bound to the reviewed evidence snapshot and never appears on the agent-facing MCP surface.
One binary. Your repositories stay yours.
macOS and Linux. No hosted account, frontend runtime, or external board required. Vikunja remains available only as an optional legacy integration.
$ curl -fsSL https://raw.githubusercontent.com/kaltstart-co/agentklar/main/install.sh | bash -s -- $ cd your-project $ agentklar init $ agentklar ui --open