OpenPocket Documentation
This documentation is organized into documentation hubs with a clear separation of onboarding, concepts, tooling, reference, and operations.
- Start with task-oriented onboarding.
- Move to concepts and execution model.
- Use tools/reference pages for exact schemas and defaults.
- Keep operations and troubleshooting separated.
All pages in this folder document implemented behavior in the current TypeScript runtime (src/).
Primary control surfaces:
- per-agent dashboard:
openpocket dashboard startor auto-started bygateway start - manager dashboard:
openpocket dashboard manager
Direction
OpenPocket is a local device-first phone-use agent aimed at real consumer scenarios, not only developer workflows.
- local execution on configurable Agent Phone targets (
emulatordefault,physical-phoneready) - one install can host multiple isolated agents
- no mandatory cloud-hosted phone runtime
- local data and permission boundary by default
- dual control model: direct human control + agent control
- remote human-auth approvals from Human Phone (one-time web link + channel fallback)
- capability-probe + agentic delegation for sensitive data flows
- optional shared relay hub + ngrok entry across many managed agents
Doc Hubs
| Hub | Purpose | Entry |
|---|---|---|
| Get Started | Install, initialize, configure, and add more agents | Quickstart |
| Concepts | Understand blueprint, architecture, and core agent mechanics | Project Blueprint |
| Tools | Skill and script authoring and runtime behavior | Skills |
| Reference | Precise schemas, defaults, formats, and commands | Config Defaults |
| Ops | Day-2 runbook and troubleshooting | Runbook |
Popular Specs
- Product blueprint: Project Blueprint
- Multi-agent setup: Multi-Agent Setup
- Prompt templates: Prompt Templates
- Default values: Config Defaults
- Session and memory formats: Session and Memory Formats
- Skill format and loading rules: Skills
- Script format and execution rules: Scripts
- CLI and channel commands: CLI and Gateway
- Remote auth and delegation design: Remote Human Authorization
- Practical multi-agent design: Recommended Multi-Agent Deployment Patterns
Scope Policy
- document implemented behavior first, and clearly mark roadmap items as planned
- mark fallback behavior and normalization rules explicitly
- keep examples executable with current CLI