One memory pool. Every AI tool. Every machine. Every teammate.

Zikra is a self-hosted shared memory layer for AI coding agents. It gives Claude Code, Gemini CLI, OpenAI Codex, ChatGPT, and any agent that can send a POST request a single, persistent, searchable memory pool — shared across tools, machines, and team members.

No cloud lock-in. No SaaS subscriptions. Your data stays on your infrastructure.

Your team uses different AI tools. None of them share memory by default.

• One developer uses Claude Code. Another uses Gemini CLI. Another uses Codex.
• Every session starts from zero. Every agent re-asks the same questions.
• Architecture decisions, error resolutions, and schema changes get re-explained dozens of times.
• Context resets mid-session and everything is lost.

Zikra fixes this. One webhook. One token. One shared context pool — automatically injected at the start of every session, automatically saved at the end.

Claude Code  ──┐
Gemini CLI   ──┼──▶  Zikra (MCP + webhook)  ──▶  PostgreSQL + pgvector
Codex CLI    ──┘
ChatGPT      ──┘          One memory pool. Every agent reads and writes it.
Any agent ────────────────┘

• Save: One POST to store a decision, error fix, schema, or requirement.
• Search: One POST for hybrid semantic + keyword retrieval — finds by meaning, not exact match.
• Auto-hooks: Stop hook fires at session end. PreCompact hook saves before context resets. You never type "save this."
• Auto-inject: Every new session gets the same shared context injected automatically.

Development teams where different people use different AI coding tools and constantly re-explain context to each agent or each other.

Solo developers who switch between Claude Code, Gemini CLI, and ChatGPT and are tired of starting from scratch every session.

Any agentic workflow where one AI agent needs to know what another decided last week, on another machine, or in another tool.

Fetch the setup guide and follow every instruction in it:

# Full installation guide
# Fetch: https://raw.githubusercontent.com/getzikra/zikra/main/prompts/g_zikra.md

Stack: PostgreSQL + pgvector + n8n + FastAPI + MCP server

Full setup typically takes 30–60 minutes. Once running, every AI agent you connect gets shared memory automatically.

Zikra organizes memory by type so agents always retrieve the right context:

┌─────────────────────────────────────────────────────────┐
│                        Zikra                            │
│                                                         │
│  FastAPI ──▶ n8n workflows ──▶ PostgreSQL + pgvector    │
│     │                                ▲                  │
│     └──────── MCP server ────────────┘                  │
│                    │                                    │
│            Role-based auth                              │
│         (owner / collaborator / guest)                  │
└─────────────────────────────────────────────────────────┘

• PostgreSQL + pgvector — persistent storage with native vector similarity search
• n8n — workflow automation, scheduled decay jobs, cron-based hygiene
• FastAPI — REST API + MCP server endpoint
• Cloudflare tunnel — optional secure public access for remote team members

• Your data stays yours. No SaaS, no third-party storage, no vendor lock-in.
• Works on-premise or on any VPS. Docker Compose gets it running in one command.
• Free forever. MIT license. No usage limits. No per-seat pricing.
• Extend it. The API is fully open. Add your own memory types, decay rules, or integrations.

shared AI memory · Claude Code memory · Gemini CLI memory · AI agent context · persistent LLM memory · self-hosted AI memory · MCP memory server · cross-agent memory · team AI context · pgvector memory · AI coding tools memory sharing · Codex shared memory · OpenAI Codex context · Claude Code context persistence