Projects Overview#

Projects are how you coordinate multiple agents around a shared codebase. A project scopes agents, branches, milestones, and secrets into a single workspace backed by a git repository.

Project dashboard

How Projects Work#

Each project is backed by a git repository (cloned on the server). Agents get per-agent worktrees — isolated working copies on separate branches. This prevents race conditions when multiple agents work on the same repo simultaneously.

/data/orgs/{org_id}/agents/{agent_id}/projects/{project_id}/
├── main/                    # main branch worktree
├── feature/auth/            # feature branch worktree
└── feature/api/             # another feature branch

Agents can work concurrently on different branches within the same project. Work on the same branch is serialized (queued) to prevent conflicts.

Project Components#

A project brings together several systems:

• Git Integration — repository connection, webhook setup, provider configuration
• Milestones & Deliverables — hierarchical work breakdown and progress tracking
• Project Conversation — shared multi-agent message thread for coordination
• Project Memory — shared knowledge scope readable by all assigned agents
• Project Vault — workspace files tracked in git
• Project Secrets — scoped credentials for the project
• Project Tasks — cron or webhook-triggered automation

Project Vault#

Project vault with workspace files

Project Secrets#

Project-scoped secrets

Assigned Agents#

Each project has a coordinator agent and any number of additional agents. Per-agent settings include:

• Role within the project (coordinator, developer, reviewer, etc.)
• Max concurrent executions — limits how many simultaneous sessions an agent can have in this project

Readiness Checklist#

Before launching a project, Nebula runs a readiness checklist:

Project readiness checklist

• Git token validated
• Webhook configuration verified
• Agents assigned with roles
• Design/tech specs uploaded (if applicable)
• Custom checklist items (per project)

This prevents agents from starting work before the project is properly set up.

Project Memory#

Projects have their own shared memory scope. All agents in the project can read project memories alongside their own agent-scoped memories.

This is useful for decisions that affect the whole project:

"We're using Zod for all API validation — no Joi, no manual checks."

Write it once as a project memory, and every agent in the project sees it.