Pull request reviews
Review on open, ready, reopen, or push with repository policy, bounded findings, feedback, and validation evidence.
Self-hosted agentic engineering
Your control plane for agentic engineering.
Bring engineering work in from GitHub, Slack, or Linear. Tasuku gives coding agents bounded context and isolated execution, then keeps approval, credentials, durable state, and publication under your control.
Runs on your infrastructure · agents never hold publication credentials
work item / implementation · #418processing
Work request01 · GitHub · Slack · Linearverified
Durable orchestration02 · work item · queue · historyqueued
Isolated agent03 · Codex or Claude · bounded workspacerunning
Control boundary04 · approval · validation · publicationpublished
Core workflows
From request to accountable result.
Work can begin in GitHub, Slack, or Linear. Tasuku records the request, runs the right agent workflow, preserves the evidence, and publishes only through a trusted integration.
Plans and implementation
Investigate a ticket, prepare a reviewable plan, wait for approval, then open a draft or ready pull request.
Slack work and research
Ask questions, request source-backed research, create implementation work, and follow the result from one thread.
Linear ticket workflows
Use Linear issues as ticket sources while repository policy keeps GitHub code targets explicit.
Recurring workflows
Schedule questions, research, and approved builds with durable occurrences, retries, and Slack delivery.
Pull request automation
Handle trusted change requests and generate structured descriptions from repository templates.
Human control
Autonomy with hard edges.
Repository enablement, role-based access, encrypted credentials, approval gates, and audit history stay outside the agent workspace.
Trusted publication boundary
Agents produce bounded artifacts. Tasuku—not the sandbox—holds integration credentials and decides what becomes a review, plan, result, or pull request.
Explicit approval
Implementation waits for an explicit human approval before any code task starts.
Roles and scope
User, admin, and super-user capabilities govern instance configuration and access.
Encrypted credentials
Provider, repository, sandbox, and MCP secrets are encrypted at rest and resolved by scope.
Auditable history
Inspect workflow runs, events, agent tasks, findings, feedback, and publication outcomes.
Runtime choices
Fit the agents to your infrastructure.
Choose the model runner and isolation boundary, then enrich the workspace with controlled context—without changing the workflow contract.
Agent providers
Configure Codex or Claude at the instance level and override per repository when needed.
Sandbox providers
Run locally with Docker or use remote Daytona and AWS Lambda execution.
Context layers
Add HTTPS MCP servers, packaged skills, custom instructions, and approved repository context.
Evidence and learning
See what happened. Improve what happens next.
Tasuku keeps operational evidence next to each work item so teams can inspect decisions, separate useful findings from noise, and turn approved feedback into reusable learning.
Review evidence
Explore findings, validations, analytics, feedback reasons, CSV exports, and the GitHub conversation recorded for each pull request.
Findline-level findings
Traceworkflow events
Learnapproved memories
Diagnostics
Correlate repository, delivery, workflow, task, attempt, and publication identifiers across structured logs and the dashboard.
Self-hosted architecture
Clear trust boundaries by design.
The control plane verifies requests from GitHub, Slack, and Linear and serves the dashboard. A separate worker claims durable tasks, creates isolated workspaces, and publishes verified artifacts.
Work channels
- GitHub repositories
- Slack conversations
- Linear tickets
Server + Postgres
- Dashboard and API
- Durable queue and state
- Encrypted configuration
Worker + sandbox
- Leased task execution
- Bounded artifacts
- No integration credentials
Tasuku documentation