Skip to main content

​
On-Prem Quickstart

This guide walks through the full on-prem install end-to-end. It uses the built-in memanto first-run wizard, which provisions the Moorcheh on-prem server, configures embedding + LLM providers, and writes all state for you. Time to complete: 5–10 minutes on first run (most of which is the initial Ollama model pull). Result: A fully working on-prem Memanto stack with remember, recall, and answer operating against a local Moorcheh container. Before you start, confirm you have Docker running and Python 3.10+ on PATH. See Requirements for details.

​
1. Install the CLI

Or with uv:
Verify:

​
2. Run the Wizard

Launch the interactive setup:
The first time you run the CLI with no subcommand, it prints the welcome banner and asks you to choose a backend:
Type 2 and press Enter. The wizard then:
  1. Verifies Docker is installed and the daemon is running.
  2. Installs moorcheh-client>=0.1.3 if it isn’t already.
  3. Prompts you for an embedding provider.
  4. Prompts you for an LLM provider (for answer.generate).
  5. Writes both into ~/.moorcheh/config.json.
  6. Runs moorcheh up --embedding-provider … --embedding-model … to start the on-prem stack.
  7. Waits for http://localhost:8080/health to return 200.
  8. If you chose Ollama, pulls the embedding and LLM models inside the Ollama container via docker exec ollama pull ….
  9. Saves the final state to ~/.memanto/on-prem/state.json.

​
3. Pick an Embedding Provider

The wizard asks:
If you pick OpenAI or Cohere, the wizard prompts (with hidden input) for your provider API key and validates it is non-empty.

​
4. Pick an LLM Provider

Next, the wizard asks for the LLM used by memanto answer:
The default mirrors your embedding choice — if you picked OpenAI for embeddings, OpenAI is suggested for the LLM, and the wizard reuses your API key so you don’t have to enter it twice. You can mix providers freely (e.g., Ollama embeddings + OpenAI LLM) — Memanto stores both choices independently. To change the model later, edit answer.model in ~/.memanto/on-prem/config.yaml.

​
5. Wait for the Server

After provider selection, the wizard prints:
If you picked Ollama, you’ll also see:
The initial Ollama pull downloads ~5 GB and can take several minutes on a slow connection. Subsequent runs reuse the cached models. When the wizard finishes you’ll see:

​
6. Verify the Install

Run the status dashboard:
You should see something like:
If the On-Prem Server line says ● offline, see Troubleshooting.

​
7. Try It End-to-End

The on-prem CLI is identical to the cloud CLI — no --on-prem flags anywhere. Create an agent, store a memory, then ask Memanto a question:
Expected behavior:
  • remember returns a memory_id immediately — no indexing delay.
  • recall returns the stored memory by semantic similarity, even with no keyword overlap.
  • answer calls your chosen LLM (Ollama / OpenAI / Cohere) with the recalled memories as context and prints the grounded answer.
Open the web UI to browse everything in a browser:
This starts the embedded Memanto server (on port 8000 by default) and opens the dashboard in your default browser.

​
8. Start the REST API (Optional)

If you want to drive Memanto from your own application code or external tools, start the local REST server:
The server listens on http://localhost:8000. It auto-detects your on-prem backend choice and routes all Moorcheh calls to http://localhost:8080. Interactive API docs are at http://localhost:8000/docs. All endpoints documented in the API Reference work identically on-prem — including /api/v2/agents/{id}/remember, /recall, /answer, /upload-file, and /batch-remember.

​
What Just Got Installed

Nothing else is installed system-wide. To remove the on-prem stack later, run moorcheh down (or docker compose down against the moorcheh project), uninstall moorcheh-client, and delete ~/.memanto/on-prem/.

​
Next Steps