Developers

Getting started

Zero is a Work execution environment, not a tracker. Setup means connecting the components needed to run Work: the desktop app, an AI provider, and optionally a work source.

Each component is independent. You can skip source integrations and run Work manually. You can skip the AI provider CLI and configure it later. Setup can be paused and resumed at any step without losing Work items or memory.

Work items and memory files persist across setup changes. Adding a source integration later does not affect existing Work.

curl http://localhost:7878/cases

Local API

Zero runs an HTTP API server at http://localhost:7878. All operations — creating Work items, running sessions, approving plans, attaching and verifying proof — are available via REST.

Full API reference →

Provider transport

Zero separates Work management from session execution. The API manages Work items, sessions, and proof. Session execution is handled by a provider.

Current builds: local CLI transport. The provider CLI (e.g., claude) must be installed on the same machine as Zero. Zero invokes it with a structured prompt and reads its output. This is a developer-side setup — it is not managed through the product UI and is not required for customers using managed plans.

Planned: API-based provider transport. Zero will call provider APIs directly, with no local CLI required. This is the target architecture for production deployments. The session model and API surface do not change — only the execution path changes.

See Integration architecture → for the full transport diagram and capability matrix.

Editor integration

Zero provides inline code context to editors via a local context server. This enables autocomplete and context-aware suggestions based on local code — no source CLI required, no cloud sync.

How it works:

• The Zero context server reads your local workspace files and open Work item memory
• Editor extensions query the local server for completions and context
• AI provider API calls are made directly from your machine — not proxied through Zero

Work context is optional. If no Work item is active, completions use only local code context. Source integrations are not involved in editor completions.

Editor extension support is in development. API shape is stable; bindings for VS Code and JetBrains are planned.

Memory model

Work memory is stored as markdown files in your workspace. Every Work item has a directory containing problem-definition.md, code-context.md, proofs.md, and final-classification.md.

Memory is local-first. No Work data leaves your machine without explicit export.

cat zero-memory/cases/{case-id}/final-classification.md
grep ConfirmedCodeBug zero-memory/cases/*/final-classification.md

Source integrations

Zero supports source integrations for importing Work context from external systems. Current integrations are development-mode only: they require the relevant CLI to be installed locally on the same machine as Zero.

If a CLI is not installed, Zero continues working — manual Work items are unaffected, and existing Work memory is not changed. Install the CLI when ready and the integration becomes available automatically.

See Repository integration for what Zero reads and writes from your git repository.

Quick links

• Case model reference
• Session model reference
• API reference
• Memory file structure