Origin Docs
Projects

Projects

Organize repository-backed development work in ORGN Studio. Each project connects a Git repository to tasks, agents, worktrees, and AI-assisted workflows inside a confidential environment.

A project in ORGN Studio is a repository-backed workspace where your team plans, executes, and reviews AI-assisted development work. Each project connects to a single Git repository and provides structured surfaces for tasks, agents, feature ideation, code exploration, and configuration.

Projects are the unit of organization inside a team. When you open a project, everything — tasks, worktrees, agent sessions, model settings — operates in the context of that repository.

What a project provides

Repository context

A connected Git repository with indexed codebase context. Agents and features reference the actual code structure, not generic prompts.

Task management

Structured tasks with priorities, assignees, statuses, and labels. Track work from triage through completion on a shared board.

Agentic workflows

Start coding sessions, spec ideation, and autonomous agent runs inside TDX-secured sandboxes tied to the repository.

Model configuration

Set default models and control TEE vs ZDR access for project-scoped inference through ORGN Gateway.

Project settings

Secrets, sandboxes, logging, usage tracking, and context documents — all scoped to the project.

Project surfaces

Each project exposes several workspaces accessible from the sidebar:

SurfacePurpose
DashboardActivity overview — recent tasks, worktrees, pull requests, and project health
TasksFull task board with filtering, assignment, and status workflows
AgentsStart AI-driven work from a natural language prompt — spec-first or code-first
FeaturesStructured feature ideation and PRD generation before code is written
ExplorerBrowse and search the indexed codebase
SettingsModels, secrets, sandboxes, logging, usage, and project metadata

How projects fit in ORGN

Team
 └── Project (one Git repository)
      ├── TDX Sandbox (confidential runtime — shared by all worktrees)
      ├── Tasks (project backlog — what needs to be done)
      ├── Feature specs & context docs
      ├── Settings (models, secrets, usage)
      └── Worktrees (isolated Git branches inside the sandbox)
           └── Sessions (agent chats — tool calls, diffs, terminal)

Cardinality: one project has one TDX sandbox, many tasks on the backlog, and many worktrees over time. A task can spawn multiple worktrees (different attempts at the same goal). Each worktree holds one or more agent sessions (chats at /chat/:id).

Tasks are planned at the project level. Worktrees are the execution layer — not a sibling lumped with tasks. The sandbox is the runtime layer — it belongs to the project, not to individual sessions.

A team can contain multiple projects — one per repository. Team members share access based on role, while project settings control model access, secrets, and execution configuration independently per repository.

Model access via ORGN Gateway

Projects route AI inference through ORGN Gateway. In project model settings, you configure:

  • Default code model — the primary model for chat, task execution, and code generation
  • TEE model access — hardware-attested models (near_*, phala_*) for sensitive inference
  • ZDR model access — zero data retention models (vercel_*) for routine development and frontier models

Not every model is TEE. The settings page lets you enable or restrict each execution type based on your project's sensitivity requirements. Model IDs use underscores throughout.

For programmatic Gateway access outside Studio, create sk-ollm-* API keys in the Gateway console and verify TEE requests in scanner.orgn.com.

Repository clone vs code index

Importing a repository creates a project-scoped Git clone inside the TDX sandbox — connected to GitHub via the ORGN GitHub App, not a detached copy that drifts forever.

ConcernBehavior
Git freshnessNew tasks and worktrees fetch the latest base branch before branching. Merge back via PR or push.
Code indexBuilt automatically on import (and during deploy). Agents use it for repository-wide context.
Index freshnessTracked by commit SHA. After significant changes, click Re-Index on the Dashboard or Refresh in Context settings.

The code index is agent context — not your source of truth. Git is.

Typical project workflow

  1. Connect a repository — ORGN clones it and indexes the codebase automatically.
  2. Configure models in Settings — choose a default and enable TEE or ZDR access as needed.
  3. Create tasks on the Dashboard or task board to structure the work.
  4. Start an agent session from Agents — describe what to build in plain language.
  5. Review diffs in the worktree, open a pull request, and mark tasks complete.
  6. Iterate — agents, tasks, and worktrees run in parallel without stomping each other.

Security posture

Every project workspace runs inside a TDX-secured sandbox:

  • Agent sessions execute in TDX-secured sandboxes isolated from other projects and teams.
  • TEE model inference produces attestation evidence verifiable through Scanner.
  • Secrets are scoped per project and never exposed to client-side code.
  • Usage and inference activity are reviewable within the platform.

See Code security and the Platform trust model for the full security architecture.

What's next

On this page