Understand unfamiliar code
Ask a focused question about selected files and bring source-backed findings into your main agent's context.
Explore research workflow →Managed delegation for AI development
Give your coding agent a coworker.
Delegate codebase research, reviews, and implementation through Triss. Use your preferred models and coding engines. Keep control of what ships.
An open-source CLI and MCP server for Claude Code, Codex, and your terminal.
npm install -g triss-coworker01 — workflows
Three workflows to start with
Get a second review
Review a branch or pull request, inspect each finding, and decide what needs to change.
Explore review workflow →Delegate a bounded change
Run an implementation task through a coding engine, inspect the resulting changes, and accept only what you have verified.
Explore implementation workflow →02 — how it works
Delegate the work. Keep the decisions.
Your agent defines the task
Choose the question, context, and acceptance criteria.
Triss invokes the configured route
Use the selected provider, model, and execution engine through CLI or MCP.
You receive a result to inspect
Read findings, source references, or implementation changes.
You decide what to use
Verify the evidence and changes before accepting them.
03 — why triss
Why add Triss to an agent you already use?
Keep your existing agent. Use Triss when you want a shared interface for delegated work across supported models, coding engines, and project tools.
A common interface
Use the same provider configuration through CLI and MCP.
Task-specific workflows
Read selected context, review changes, or delegate an implementation task without inventing a new prompt-and-script workflow each time.
Results you can inspect
Keep findings, changes, execution details, and usage information available for review.
If your agent's built-in tools already meet your needs, you may not need another layer. Triss is useful when you want explicit delegation and a common interface across tools.
04 — controls and boundaries
Your choices. Clear boundaries.
Models and engines
Choose the supported provider, model, and engine for the work. Check theconfiguration guide for current route requirements.
Data
Selected code and task context may be sent to your model provider. Triss has no developer telemetry. Security and data flows
Execution
A Git worktree separates changes. It is not an operating-system sandbox.Coding engines and isolation
Acceptance
A completed run is not proof that a task is correct. Review the result before using it. Implementation workflow
05 — cost
Cost is one part of the decision.
Delegating to a lower-cost model can reduce inference spend. The result depends on the task, selected model, cache usage, and any retries. Compare the assumptions, not just the headline.
06 — commands
The command overview
Browse common CLI commands with examples and flags — searchable, in one place. The complete interface contract lives in each command's--help output, theCLI reference, and theMCP reference.
07 — install
Install Triss
A local CLI with no developer telemetry. Requires Node ≥ 22.12 and a credential for your chosen provider.
npm install -g triss-coworker