Agents
Built-in Synap AI, external agents, MCP, skills, and governance model.
Synap supports two complementary agent layers:
- Built-in intelligence running with your pod (orchestrator + specialists).
- External agents connected through open protocols (Claude Code, Claude Desktop, Cursor, Raycast, OpenClaw, custom).
All mutations — from either layer — are governed by the same permission and proposal model.
Quick start
npx @synap-core/cli connect --target=claude-code
# or: --target=claude-desktop, cursor, raycast, openclawThen jump to Connect your AI for the per-client guides.
Connection protocols
| Protocol | Best for | Endpoint |
|---|---|---|
| Agent Skills | Claude Code, Claude Desktop, OpenClaw — teach the AI how to use Synap | Markdown files in the AI's skills directory |
| MCP | Claude Desktop, Cursor, custom agents — direct typed tool calls | POST /mcp |
| Hub Protocol REST | Full agent workflows, CRUD, memory, channels | /api/hub/* |
| OpenAI-compat | Reuse existing LLM clients as-is | /v1/chat/completions |
Built-in Synap AI
- Works in channels and branches.
- Can propose structured changes.
- Supports proactive nudges and briefings.
- Uses the same governance model as external agents.
Details: Synap AI.
External agents
- Connect via API keys (
hub-protocol.*,mcp.*). - Scope-limited per key.
- Writes outside the auto-approve whitelist create proposals for your review.
- No silent mutations.
Start here: Connect your AI.
The three Synap skills
Installed together on skill-capable clients:
synap— core data operations (read, write, link, search, remember)synap-schema— extend the data model (new profiles, new properties)synap-ui— build views, dashboards, bento layouts
Details: Skills.
Practical paths
- End-to-end onboarding (any AI): Connect your AI
- Message-native flows: Channels and agents
- Direct tool calls: MCP
- OpenClaw runtime: OpenClaw
Governance
- AI writes pass permission checks.
- Non-auto-approved operations create proposals.
- Full audit trail remains in event history.