Skip to main content

Changelog

v2.2.0 — Current​

New: Chat Endpoint (RAG + LLM)

  • POST /agent/memory/chat — full RAG pipeline endpoint. Retrieves relevant memories, reranks them, and generates a natural language answer with citations using Gemini. Supports stream=true for SSE.
  • Extracted shared chatPipeline service — both dashboard (/memory/query) and agent (/agent/memory/chat) use the same retrieval + generation pipeline.
  • /memory/query refactored from ~1275 to ~270 lines.

CLI Rename: smem → sm

  • CLI binary renamed from smem to sm for brevity.
  • All commands updated: sm add, sm search, sm ask, sm profile, sm status, sm agents.
  • sm ask now calls /agent/memory/chat and displays LLM answers with sources and citations.

SDK Updates

  • @sharedmemory/sdk — added chat(query, opts?) method returning ChatResult { answer, sources, citations }.
  • sharedmemory-ai (Python) — added chat(query, **opts) to both sync and async clients.
  • @sharedmemory/mcp-server — updated recall description to "Use this to retrieve context BEFORE answering".

Docs parity

  • Refreshed quickstart plus SDK/MCP/CLI references so examples prefer Settings → API Keys (sm_org_*/sm_proj_*) while documenting agent (sm_agent_*), CLI OAuth (sm_live_*), and Remote MCP bearer tokens (sm_oauth_*).
  • MCP install helper snippets now mirror the published README wording.

v2.1.1​

Bug Fixes

  • Fixed password validation mismatch: login page now enforces the same 8-character minimum as registration
  • Reset password page now validates minimum length and includes a confirm-password field
  • Volume name-uniqueness race condition resolved with PostgreSQL advisory locks inside a transaction
  • Export filename header injection hardened — Content-Disposition volume IDs are now sanitized

API Improvements

  • Agent memory single-item CRUD: GET/DELETE/PATCH /agent/memory/:memoryId
  • Batch operations: POST /agent/memory/delete/batch and /agent/memory/update/batch with Zod validation
  • Webhook delivery: 3 retries with exponential backoff (1 s / 5 s / 15 s), skips retry on 4xx
  • Webhooks auto-disabled after 10 consecutive delivery failures; error state cleared on success

SDKs

  • @sharedmemory/sdk v2.1.0 — added startSession, endSession, getSession, listSessions, exportMemories, importMemories, extract, createExtractionSchema, listExtractionSchemas
  • sharedmemory-ai (Python) v0.3.0 — added create_agent, list_agents, get_agent, update_agent, delete_agent, rotate_agent_key, list_orgs, get_org, list_org_members, apply_promo (sync + async)

v2.1.0​

Data Model

  • Organizations as top-level entities (auto-created on signup)
  • Projects (volumes) scoped to organizations
  • Agent profiles with system prompts, scoped to projects
  • Scoped API keys: sm_org_rw_, sm_proj_rw_, sm_proj_r_, sm_agent_
  • Key rotation without recreating agents

Memory Pipeline

  • Sync path: embed → guard → PG + Qdrant (sub-400ms)
  • Async path: BullMQ → Neo4j knowledge extraction + contradiction detection
  • Separated sync/async for predictable latency

SDKs & Tools

  • @sharedmemory/sdk v2.0.0 — agents and orgs namespaces
  • @sharedmemory/cli v2.0.0 — smem agents list|create|delete|rotate-key
  • @sharedmemory/mcp-server v2.0.0 — manage_memory tool, updated project terminology

Dashboard

  • New navigation: Memory → Agents → Platform
  • Agent management UI (create, edit, delete, key rotation)
  • Organization settings (members, promo codes, plan info)
  • Removed 3D graph mode (2D only for performance)

v2.0.0​

Architecture

  • Knowledge graph (Neo4j) with Entity, Fact, Summary, and Document nodes
  • Multi-model LLM support with per-volume agent configuration
  • BullMQ job queue (Redis-backed) replacing in-memory queue
  • Structured logging with Pino + GKE Cloud Logging severity mapping

Memory Pipeline

  • Classify → Guard → Extract → Graph → Embed → Store
  • Conflict detection and memory merging
  • Configurable guard system with confidence scoring
  • Entity and relationship extraction via LLM

Security

  • Zod validation on all agent API inputs
  • Rate limiting: 30/min write, 60/min query, 50/hr upload
  • Volume access control (agents must be connected to volumes)
  • SHA-256 hashed API keys (only prefix stored for display)

SDKs & Tools

  • @sharedmemory/sdk — TypeScript/JavaScript client
  • @sharedmemory/cli — Terminal interface (smem)
  • @sharedmemory/mcp-server — MCP for Claude, Cursor, Windsurf, VS Code

Integrations

  • GitHub (issues, PRs, commits)
  • Slack (channels, threads)
  • Notion (pages, databases)
  • Linear (issues, projects, cycles)

Dashboard

  • Interactive 2D knowledge graph visualization
  • Document library with upload, toggle, and delete
  • Chat with multi-turn conversation history
  • Real-time graph updates (WebSocket + polling)
  • User identity resolution in queries

API

  • Agent registration with auto-volume connection
  • Memory write, query, and webhook subscription
  • Entity lookup, search, and graph exploration
  • Document upload and ingestion
  • Inter-agent broadcast via WebSocket

v1.0.0​

  • Initial release with basic memory storage and retrieval
  • Single-volume support
  • Simple vector search (Qdrant)