Recalla

Switch devices, keep coding. Your AI remembers everything.

Recalla is an open-source, self-hosted AI memory infrastructure. It solves the context loss problem when developers switch between devices — your AI knows what you were working on, which branch, what decisions were made, and continues seamlessly.

The Problem

You code on your MacBook, switch to your desktop — AI has no idea what you did
Every new session starts from zero, you repeat yourself
Wrong deployments, forgotten decisions, duplicated code

How It Works

Any Device → AI Tool (Claude/Cursor/Codex) → MCP Protocol → Recalla Server
                                                                 ↓
                                              Postgres + Qdrant (Vector Search)
                                                                 ↓
                                              Memory stored, synced, searchable

Recalla sits between your AI tools and a persistent memory layer. It captures context, stores it on YOUR server, and brings it back when you need it — on any device.

Features

Cross-device memory sync — Save AI sessions, decisions, code context to your own server
Semantic search — Find relevant memories using natural language (powered by vector search)
AI session compression — Compress long conversations into structured summaries
Auto-tagging — Memories automatically tagged based on content (decision/bug/deploy/code...)
Importance scoring — Memories scored 0.1-1.0, decisions auto-scored higher
Knowledge graph — Entity facts with temporal validity (who works on what, when)
Contradiction detection — Automatically find conflicting memories
Team collaboration — Create teams, invite members, share memories, team-wide search
Config sync — Push CLAUDE.md and AI tool configs to server, new devices pull automatically
MCP protocol — One server works with Claude Code, Cursor, VS Code, Codex (12 tools)
GitHub integration — Auto-record pushes and PRs as memories via Webhook
Telegram bot — Search, save, ask questions, get briefings from Telegram
CLI tool — recalla search/save/context/status from terminal
Multi-provider AI — Support OpenAI, Qwen, DeepSeek, Ollama, or any OpenAI-compatible API
Dashboard — Web UI to manage memories, sessions, projects, teams, and settings
Daily briefing — Generate project activity summaries
User auth — Login system with API key management
Self-hosted — Your data stays on your server, zero third-party dependency
i18n — English and Chinese

Quick Start

git clone https://github.com/gentpan/recalla.git
cd recalla
cp .env.example .env
docker compose up -d

Open http://your-server:14200 — default login: admin / admin123

Connect Your AI Tools

Claude Code

claude mcp add-json recalla '{"type":"http","url":"https://your-server/mcp"}' --scope user

Cursor

Settings → MCP → Add Server → URL: https://your-server/mcp

Codex (OpenAI CLI)

# ~/.codex/config.toml
[mcp.recalla]
type = "url"
url  = "https://your-server/mcp"

VS Code

{
  "mcp": {
    "servers": {
      "recalla": {
        "url": "https://your-server/mcp",
        "type": "streamableHttp"
      }
    }
  }
}

Auto-recall Rules

Add to your project's CLAUDE.md / AGENTS.md / .cursorrules:

## Recalla Rules
- On conversation start: call context_restore
- On important decisions: call memory_save
- On conversation end: call session_sync
- Before new tasks: call memory_search
- After config changes: call config_push

MCP Tools (12)

Tool	Description
`memory_save`	Save a memory (auto-tagged, auto-scored)
`memory_search`	Semantic search across all memories
`context_restore`	Restore project context (last device, branch, recent work)
`session_sync`	Sync current AI session to cloud
`session_compress`	AI-compress long sessions (auto-saves as memory)
`project_list`	List all recorded projects
`config_push`	Push local AI config files to server
`config_pull`	Pull latest config from server
`team_search`	Search across all team members' memories
`team_share`	Share a memory with your team
`add_fact`	Add entity fact to knowledge graph
`query_facts`	Query knowledge graph facts

REST API

All endpoints require Authorization: Bearer <api-key> header. Generate API keys in Dashboard → Account.

Memory

Method	Endpoint	Description
POST	`/api/memory/save`	Save memory (auto-tagged)
POST	`/api/memory/search`	Semantic search
PUT	`/api/memory/{id}`	Update memory
DELETE	`/api/memory/{id}`	Delete memory
POST	`/api/memory/check`	Contradiction detection

Context & Sessions

Method	Endpoint	Description
POST	`/api/context/restore`	Restore project context
POST	`/api/session/sync`	Sync session
POST	`/api/session/compress`	Compress session
GET	`/api/sessions`	List sessions
POST	`/api/sessions/import`	Bulk import sessions

Projects

Method	Endpoint	Description
GET	`/api/projects`	List projects
GET	`/api/project/{name}`	Project detail
GET	`/api/project/{name}/memories`	Project memories
GET	`/api/project/{name}/sessions`	Project sessions
GET	`/api/stats`	Dashboard stats

Teams

Method	Endpoint	Description
POST	`/api/teams`	Create team
GET	`/api/teams`	List my teams
GET	`/api/teams/{id}/detail`	Team detail (members, projects, activity)
POST	`/api/teams/{id}/invite`	Invite member by username
POST	`/api/teams/{id}/projects`	Link project to team
POST	`/api/teams/{id}/search`	Search team memories
POST	`/api/teams/{id}/share`	Share memory to team
GET	`/api/invites`	Pending invites
POST	`/api/invites/{id}/accept`	Accept invite

Knowledge Graph

Method	Endpoint	Description
POST	`/api/kg/facts`	Add entity fact
GET	`/api/kg/facts`	Query facts (?subject=&predicate=&project=)
DELETE	`/api/kg/facts/{id}`	Invalidate fact

Config Sync

Method	Endpoint	Description
POST	`/api/config/push`	Push config to server
POST	`/api/config/pull`	Pull config from server
GET	`/api/config/list`	List synced configs

Briefing

Method	Endpoint	Description
POST	`/api/briefing`	Generate daily briefing

GitHub

Method	Endpoint	Description
GET	`/api/github/repos`	List GitHub repos
GET	`/api/github/repos/{owner}/{repo}/commits`	Repo commits
POST	`/api/github/webhook`	GitHub webhook receiver

Auth

Method	Endpoint	Description
POST	`/api/auth/login`	Login
GET	`/api/auth/me`	Current user
POST	`/api/auth/password`	Change password
POST	`/api/auth/username`	Change username
GET	`/api/auth/keys`	List API keys
POST	`/api/auth/keys`	Generate API key
DELETE	`/api/auth/keys/{id}`	Delete API key

Architecture

┌─────────────────────────────────┐
│  Claude / Cursor / Codex / VSC  │
│         (Any Device)            │
└──────────┬──────────────────────┘
           │ MCP Protocol (12 tools)
           ↓
┌─────────────────────────────────┐
│       Recalla Server (Go)       │
│  REST API + MCP + Dashboard     │
│  + Telegram Bot + CLI           │
└──────┬─────────┬────────┬───────┘
       ↓         ↓        ↓
   Postgres   Qdrant   AI API
   (metadata) (vectors) (embed/compress)

Tech Stack

Go — API server (standard library, no framework)
PostgreSQL — Structured storage (users, memories, sessions, teams, knowledge graph)
Qdrant — Vector search for semantic memory retrieval
MCP — Model Context Protocol (Anthropic standard)
Docker — One-command deployment

AI Provider Support

Configure in Dashboard → Settings. Multiple providers simultaneously.

Provider	Embedding	LLM (Compression)	Note
OpenAI	text-embedding-3-small/large	gpt-4o-mini	Best quality
Qwen	text-embedding-v3	qwen-plus	Good for Chinese, affordable
DeepSeek	Not supported	deepseek-chat	LLM only
Ollama	nomic-embed-text	Local models	Self-hosted

Telegram Bot

Commands: /search /save /ask /context /projects /compress /briefing /help

Setup: Create bot via @BotFather → Set RECALLA_TELEGRAM_TOKEN → Set webhook to https://your-server/api/telegram/webhook

CLI Tool

go install github.com/gentpan/recalla/cmd/recalla@latest

export RECALLA_URL=https://your-server
export RECALLA_KEY=rk-xxx

recalla search "login bug"
recalla save "Chose PostgreSQL for JSON support"
recalla context zhanxing.io
recalla status
recalla push claude-md ~/.claude/CLAUDE.md
recalla pull claude-md

Development

go run ./cmd/server/           # Run server
go run ./cmd/recalla/ help     # CLI help
go build ./cmd/server/         # Build server
go build ./cmd/recalla/        # Build CLI
GOOS=linux GOARCH=amd64 go build -o dist/recalla-server ./cmd/server/  # Cross-compile

License

MIT

Name		Name	Last commit message	Last commit date
Latest commit History 14 Commits
cmd		cmd
internal		internal
migrations		migrations
web/static		web/static
.env.example		.env.example
.gitignore		.gitignore
Dockerfile		Dockerfile
LICENSE		LICENSE
Makefile		Makefile
README.md		README.md
docker-compose.yml		docker-compose.yml
go.mod		go.mod
go.sum		go.sum
recalla-claude.md		recalla-claude.md
recalla-rules.md		recalla-rules.md

Folders and files

Latest commit

History

Repository files navigation

Recalla

The Problem

How It Works

Features

Quick Start

Connect Your AI Tools

Claude Code

Cursor

Codex (OpenAI CLI)

VS Code

Auto-recall Rules

MCP Tools (12)

REST API

Memory

Context & Sessions

Projects

Teams

Knowledge Graph

Config Sync

Briefing

GitHub

Auth

Architecture

Tech Stack

AI Provider Support

Telegram Bot

CLI Tool

Development

License

About

Topics

Resources

License

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages