Deploy and Host AI Memory Server on Railway

ai-memory is a long-term memory server for coding agents. Claude Code, Codex, OpenCode, Gemini CLI, Cursor and others send their sessions to it over MCP and lifecycle hooks. The next session, on any machine and with any agent, starts with a handoff of where you left off and can search everything you did before. It's one Rust binary with SQLite full-text search and a git-versioned markdown wiki.

About Hosting AI Memory Server

It's one service and a Railway volume.

Locked from the first boot. A random bearer token is generated at deploy time. Every MCP, hook and handoff request must carry it; anything else gets a 401.
One server for all your machines. Your laptop, desktop, cloud agent boxes and teammates point at the same HTTPS address, so memory follows you instead of living in one home directory.
Survives redeploys. The SQLite database and the markdown wiki (with its git history) live on the volume.
Small. The image is about 50 MB and the server uses about 200 MB of RAM, so it fits the Free plan.

Common Use Cases

Picking up a Claude Code session on another computer, or in Codex, without re-explaining the project
A shared memory for a team's agents: decisions, gotchas and procedures one person's agent learned are found by everyone else's
Handing work between agent vendors (Claude Code, Codex, OpenCode, Gemini CLI, Kimi, Copilot CLI and more) through one MCP server

Dependencies for AI Memory Server Hosting

Nothing else: the database is SQLite on the service's volume
Optional: an LLM key (Anthropic, OpenAI, Gemini, OpenRouter or any OpenAI-compatible endpoint) for better session summaries and semantic search

Deployment Dependencies

Implementation Details

Connect an agent. Install the ai-memory CLI on your machine (see the installation guide: Homebrew, release tarball or the Docker wrapper), then point it at your Railway domain with the AI_MEMORY_AUTH_TOKEN value from the Variables tab:

ai-memory install-mcp   --client claude-code --apply \
    --server-url "https://.up.railway.app/mcp" --auth-token ""
ai-memory install-hooks --agent  claude-code --apply \
    --server-url "https://.up.railway.app" --auth-token ""

Repeat with --client codex / --agent codex, opencode, gemini-cli, cursor and so on for each agent. Any other MCP client can use the streamable HTTP endpoint https://.up.railway.app/mcp with an Authorization: Bearer header.

Teammates. The generated token is the root credential. For a team, give each person their own key with ai-memory user / ai-memory api-key (see docs/users.md upstream) so writes are attributed and keys can be revoked one at a time.

Better summaries and search. Without an LLM, ai-memory uses full-text search and rule-based session summaries, which already work. To add LLM consolidation, set for example AI_MEMORY_LLM_PROVIDER=anthropic and ANTHROPIC_API_KEY, or AI_MEMORY_LLM_PROVIDER=openai-compat with AI_MEMORY_LLM_BASE_URL=https://openrouter.ai/api/v1, AI_MEMORY_LLM_MODEL and LLM_API_KEY. For hybrid vector search add AI_MEMORY_EMBEDDING_PROVIDER=openai and OPENAI_API_KEY.

Custom domain. Add its hostname to AI_MEMORY_ALLOWED_HOSTS (comma separated). The server refuses requests for hostnames not on that list, which protects it from DNS-rebinding attacks.

Private network. Agents running in the same Railway project can use http://${{AI Memory.RAILWAY_PRIVATE_DOMAIN}}:49374 with the same token. That traffic isn't billed as egress.

Backups. Everything is in the volume at /data: wiki/ is a git repository you can push to a private remote, and db/ holds the SQLite indexes (see docs/backup.md upstream).

Why Deploy AI Memory Server on Railway?

Railway is a singular platform to deploy your infrastructure stack. Railway will host your infrastructure so you don't have to deal with configuration, while allowing you to vertically and horizontally scale it.

By deploying AI Memory Server on Railway, you are one step closer to supporting a complete full-stack application with minimal burden. Host your servers, databases, AI agents, and more on Railway.