diraigent

coordinate your agents.
build together.

a self-hosted platform for AI-assisted software development. Plan work, run coding agents in isolated worktrees, and review results together from the web or iOS app.

self-hosted shared workspaces task orchestration your choice of agent human-in-the-loop
Diraigent dashboard

one workspace for the whole project

plan with context

Keep tasks, acceptance criteria, decisions and project knowledge together. Give each task the context needed to start work.

choose your agents

Use OpenCode by default, or configure Codex or Claude Code. Repository instructions and skills guide the selected agent.

work together

Share projects with viewers, editors and managers. Review task results and give demo users read-only access.

from task to delivery

execution

tasks with clear ownership and review

Agents claim ready tasks, work in isolated Git worktrees, and report results. Use repository guidance and skills for the work, and request human review when needed. Orchestra coordinates execution and delivery.

Task board
goals

turn goals into actionable tasks

Describe a goal and use project chat to plan concrete tasks with acceptance criteria. Track epics, features and milestones on the board, then make tasks ready for authorized workers.

Goal decomposition view
review

humans decide what matters

Merge conflicts, ambiguous requirements, quality gate failures — these surface in a review queue for human decision. Approve, rework, or resolve with full context. Agents handle the routine; you handle the judgment calls.

Review queue view
knowledge

your project gets smarter over time

Knowledge entries, architectural decisions, and observations accumulate as agents work. Keep relevant context available for later tasks, agents and sessions.

Project knowledge view

different agents, different permissions

Define agent roles with scoped authority for execution, review and project management. Human project permissions are managed separately, with viewer, editor and manager access.

releases are a first-class concept

Choose a project Git delivery strategy for merging completed work or retaining branches. Release tools support merge strategies and tagging when you are ready to ship.

a lightweight control plane, durable workers

keep coordination central and detailed project content with your Orchestra.

api and shared state

The API owns sign-in, permissions and task coordination. PostgreSQL keeps shared state and references to content stored on the assigned worker.

Orchestra on your machine

Run Orchestra on a host or in a container. It manages repositories, task worktrees and execution. Enable full storage to keep detailed project content and content search on the worker.

web and iOS

Manage work and chat through the authenticated API. Project permissions apply to both clients; workspace and access management are available on the web.

Full storage requires the assigned Orchestra to be online for detailed reads. Persist its data directory and back it up together with the central database.

from goal to shipped code

1

define a goal

Describe what you want at any level: an epic, a feature, a bug fix. Attach acceptance criteria.

2

plan concrete tasks

Use project chat to refine the work into tasks. Add acceptance criteria and make tasks ready when they can be executed.

3

agents execute in isolation

Each task runs in its own git worktree. Multiple agents work in parallel with full logs and structured output.

4

review what needs judgment

Blockers, conflicts, and quality gate failures surface in your review queue. Approve, rework, or resolve.

5

merge, release, learn

Completed work merges through your configured strategy. Knowledge and decisions feed back into the project for the next cycle.

built for reliability

rust backend

The Rust API coordinates shared state and authorized content access. Orchestra handles tools, repositories and task execution on its host.

self-hosted

Host Diraigent and its durable storage on your infrastructure. Your chosen AI provider and MCP tools determine where prompts and tool data are processed.

plain docker setup

Use Docker Compose for the API and web app, and run Orchestra in a container or directly on a development machine with the tools your project needs.

run on your infrastructure

run diraigent on your own infrastructure.
uses OpenCode by default, with Codex and Claude Code optional. Orchestra runs the selected agent in an isolated git worktree for each task.

Requires Git, Docker Compose, Bash, curl and Python 3 on Linux or macOS. Set a provider API key or use the OpenCode login helper, then choose an explicit provider/model. Worker content and provider accounts use persistent volumes.


git clone https://github.com/diraigent/diraigent.git
cd diraigent/startup
cp .env.example .env    # edit .env for your setup
chmod +x start.sh
./start.sh --build --login  # optional: build and connect a provider
./start.sh --models     # choose OPENCODE_MODEL in .env
./start.sh --build      # builds images, registers agent, starts services
      

after startup

  1. create a project — point it at your git repo and its default branch
  2. chat with the assistant — verify the configured agent responds
  3. create a task — fill in spec and acceptance criteria
  4. make the task ready — an authorized, connected Orchestra picks it up
git credentials

the orchestra pushes branches and merges results back to your remote. for this to work, git must be able to authenticate inside the container — mount a .netrc file, an ssh key, or configure a credential helper. without credentials, agents work locally but push/merge to remote will fail.