@arvoretech/pi-memory

Cloud-based team memory for Pi. All intelligence (embeddings, chunking, ADD/UPDATE/NOOP decision, PII sanitization) runs server-side in api-arvore. The extension only sends raw text.

How it works

1. /memory-login → GitHub OAuth via api-arvore (org-based access: arvoreeducacao or SuperAutor).
2. During the session → on each turn_end (debounced) the new turns are sent to api-arvore, which sanitizes, embeds, and decides ADD/UPDATE/NOOP. Final flush on session_shutdown.
3. On before_agent_start → the current prompt is used to retrieve relevant team memory, injected into the system prompt.

Memory is a single org-wide pool (cross-user). Two tiers: raw (episodic chat, decays over time) and curated (promoted, never auto-deleted).

Commands

| Command | Description | |---------|-------------| | /memory-login | GitHub OAuth login | | /memory-logout | Clear credentials | | /memory-incognito | Toggle incognito (skip saving this session) | | /memory-status | Show auth/config info | | /memory-search <query> | Manual semantic search | | /memory-candidates | List raw memories eligible for promotion | | /memory-promote <id> | Promote a raw memory to curated | | /memory-curate <category>\|<title>\|<content> | Create a curated memory directly | | /memory-forget | Delete the current session from memory (retroactive incognito) |

Config

• PI_MEMORY_API_URL — override the api-arvore base URL (default: production).

Dependencies

None at runtime — uses native fetch and Node builtins. Backend: api-arvore pi-memory module.