FAQ

Common questions and answers. If you don't find what you need here, open an issue on GitHub.

Is EXMER hosted?

No. You run it on your own server. The reference deployment is at exer.me, but that's just the author's instance. Clone the repo and deploy your own.

Is it free?

Yes. MIT licensed. No paid tiers, no hosted service, no telemetry.

Do you store my data?

No. Everything lives on your server. SSH credentials are encrypted locally with AES-256-GCM keyed from JWT_SECRET. Agent configs, sessions, and RAG data live on the target OpenClaw servers you connect to. Nothing gets sent to the EXMER author or any third party.

What if I lose my JWT_SECRET?

All stored SSH credentials become unreadable and you have to re-enter them from the UI. Agent configs and RAG data on OpenClaw servers are unaffected — they live outside EXMER. Back up your .env file.

Does AI Analysis send my server data to Anthropic?

Yes — that's how Claude works. The diagnostic snapshot is sent as part of the API request. If you're not comfortable with that, don't set ANTHROPIC_API_KEY and AI Analysis will be disabled. Nothing else depends on it.

Why SQLite instead of Postgres?

EXMER is designed for single-digit to low-double-digit server counts. SQLite with WAL mode handles that load with zero operational overhead. If you actually need Postgres, you're running a different product than what EXMER aims to be.

Can I use EXMER without Telegram?

No. EXMER is specifically a Telegram Mini App — auth, notifications, and deployment all assume Telegram is the user's interface. If you want a web-only version, that's a different project.

Does it work on iOS and Android?

Yes. The Mini App runs inside Telegram on both platforms. We test primarily on Android (Samsung Galaxy S-series). Known quirks: Android lacks a native browser back button, so every page has an explicit Back button in the header.

Can I run multiple instances of EXMER?

Yes. Each instance has its own database, JWT secret, and bot. You can run dev + staging + prod with three bots.

What about subscriptions / billing?

Not built-in. See Monetization for models we've considered but not implemented. If you want to add one, the server_access and audit_log tables are good hooks.

Why Voyage AI for embeddings, not Anthropic?

Anthropic doesn't make embedding models. They recommend Voyage as a separate service (Voyage is their official embedding partner). Voyage keys start with pa-. Don't paste your Anthropic key into the Voyage field.

Why not use the official Ollama installer?

Because it's a black box that breaks when upstream changes asset names. We hit a real incident where curl | sh hung for 3+ minutes downloading a tarball from a URL that returned 404 for one of the asset variants. Our 9-step manual install handles the same cases explicitly and reports progress at each step.

Can agents read each other's RAG memory?

No. Each agent has its own query.py with AGENT_ID hardcoded into the file. There is no global script with a --agent flag. Plus chmod 700 on rag dirs — though this only helps if OpenClaw runs agents as separate UNIX users (it often doesn't). See RAG memory.

My agent says it has RAG but there's no data

Classic hallucination. Check the RAG Manager — if the badge is settings only instead of active, the agent has nothing to search and is making it up. Either upload and reindex documents, or disconnect RAG from that agent so it stops claiming to have memory.

How do I back up my EXMER database?

Automatic daily backups are built in — see Security. Super-admins can also trigger a manual backup at any time via the admin panel.