Platform Architecture
A unified API for agents, skills, and resources — built on Supabase PostgreSQL with multi-tenant org scoping.
Skill Abstraction Levels
Skills are the runtime foundation. They form a hierarchical ontology from atomic actions to meta-meta-strategies, composed into DAGs with typed edges.
Single executable actions — the leaf nodes of the skill tree.
send-email, parse-json, read-file
Orchestrated sequences of atomic skills with defined ordering.
research-topic, code-review
Strategies that select and compose lower-level skills at runtime.
project-planning, debugging-strategy
Arbitrarily deep abstraction — meta-strategies over meta-strategies.
curriculum-design, org-optimization
Agent Lifecycle
Each user can register up to 16 agents per org. Agent types include: ai_agent, cli_tool, iot_device, mobile_app, and browser_ext.
API Endpoint Groups
All endpoints follow Google Workspace CLI conventions: pageSize/pageToken pagination, filter expressions, field masks, and orderBy.
/api/v1/agents
- GET /agents — list agents
- POST /agents — register agent
- POST /agents/{id}/bind-key — bind API key
- POST /agents/{id}/heartbeat — agent heartbeat
- GET /agents/{id}/config — resolved bootstrap config
/api/v1/skills
- GET /skills — list skill definitions
- POST /skills — create skill
- GET /skills/{id}/tree — full decomposition tree
- GET /skills/{id}/ancestors — path to root
- CRUD /skills/{id}/compositions — manage DAG edges
- CRUD /skills/{id}/implementations — executable bindings
/api/v1/resources
- GET /resource-types — domain type ontology
- GET /resources — list digital twin instances
- CRUD /resources/{id}/edges — relationship graph
- CRUD /agents/{id}/bindings — agent-resource access
- GET /agents/{id}/skills/resolve — full runtime skill tree
/api/v1/me
- GET /me — user profile
- GET /me/orgs — my organizations
- GET /me/agents — my agents across orgs
- GET /orgs — list organizations
- GET /orgs/{slug}/members — org members