Recalla is an open-source, self-hosted memory infrastructure for AI systems. Store, search, and sync context across models, tools, and devices.
Switch from MacBook to desktop — AI has no idea what you were working on. Decisions forgotten, progress lost.
Explain the same architecture, same codebase, same constraints — every single session, every single device.
Token limits, session breaks, tool switching. Context is a short-term cache that vanishes instantly.
Your memories live on your server. Switch devices, AI picks up exactly where you left off. Branch, device, decisions — all remembered.
Find relevant context with natural language. Powered by vector search (Qdrant). Supports OpenAI, Qwen, Ollama embeddings.
One server, all tools. Works with Claude Code Cursor VS Code Codex via standard MCP.
Long sessions compressed into structured summaries. Key decisions, code changes, and TODOs preserved. Context limits handled.
Auto-record pushes and PR events as memories via webhook. Browse repos and commits from dashboard.
Create teams, invite members by username, share memories across the team. Team-wide semantic search finds knowledge from any member.
Entity facts with temporal validity. Track who works on what, which tech stack is used, when decisions were made. Contradiction detection built-in.
Memories automatically tagged (decision/bug/deploy/code) and scored by importance. Decisions ranked higher. Zero manual effort.
Push CLAUDE.md and AI configs to server. New device pulls the latest version automatically. No Syncthing needed.
Search memories, ask AI questions, get daily briefings — all from Telegram or terminal. recalla search "login bug"
Your data on your server. Zero third-party dependency. Docker one-command deploy. MIT licensed.
| Tool | Description | When to use |
|---|---|---|
| context_restore | Restore project context — last device, branch, recent memories | Every conversation start |
| memory_save | Save a memory — decisions, code changes, bug fixes, deployments | After important actions |
| memory_search | Semantic search across all memories | Before starting new tasks |
| session_sync | Sync current session to cloud with device and branch info | End of conversation |
| session_compress | AI-compress long sessions into structured summaries | When context gets long |
| project_list | List all recorded projects | Discovery |
| config_push | Push local AI config files to server | After config changes |
| config_pull | Pull latest config from server to local | New device setup |
| team_search | Search across all team members' memories | Team collaboration |
| team_share | Share a memory with your team | Knowledge sharing |
| add_fact | Add entity fact to knowledge graph | Recording facts |
| query_facts | Query knowledge graph with temporal validity | Who works on what |
Clone, deploy, connect. Your AI remembers everything from now on.