pi-backend-agent-runtime

name	pi-backend-agent-runtime
description	Use this skill when executing tenant chat, ingestion, memory reads, skill routing, or API calls on behalf of a tenant. Pi is the backend agent runtime. Trigger when: a tenant message arrives, a document needs ingesting, a skill needs routing, memory needs loading, or any API call targets /v1/tenants/*.

Pi Backend Agent Runtime

Role

Pi owns all tenant-facing AI execution. When a user sends a chat message through the frontend, the website POSTs to Pi. Pi loads tenant memory, selects relevant skills, executes, and returns a response.

Execution Loop

1. Receive request (chat, ingest, lead, etc.)
2. Load tenant memory context (namespaced by slug)
3. Select relevant skills based on intent
4. Execute skill(s) with tenant context injected
5. Return response + memory token count
6. Write audit log entry

Runtime Requirements

• CLI mode: factoryctl commands
• API mode: Pi REST API (pi/api/openapi.yaml)
• Memory: isolated per tenant, survives suspension
• Skill routing: declarative via skills/*/SKILL.md frontmatter

Never

• Cross-tenant memory reads
• Raw LLM calls without tenant context injection
• Executing billing, legal, or destructive actions without a human approval gate

References

• pi/AGENTS.md
• pi/api/openapi.yaml
• pi/memory/tenant-memory-contract.md
• Pi upstream