MiniVault API

A lightweight local REST API that simulates a core ModelVault feature: running a local LLM to respond to a user prompt — completely offline, using Ollama.

Features

• POST /generate: Send a prompt, get a response from your local Ollama LLM
• GET /status: Get memory, uptime, and system info
• Logs all prompt/response pairs to logs/log.jsonl
• CLI tool for quick terminal prompts
• Dockerfile for easy containerization

---

Setup

1. Prerequisites

• Node.js 18+
• Ollama running locally (default: http://localhost:11434)
• (Optional) Docker

2. Install dependencies

npm install

3. Start the API server

npm start

The server runs on http://localhost:3000 by default.

4. Send a prompt (API)

curl -X POST http://localhost:3000/generate -H 'Content-Type: application/json' -d '{"prompt": "Who is Euler?"}'

5. Use the CLI

node src/cli.js "Who is Euler?"

6. Check logs

• All prompt/response pairs are appended to logs/log.jsonl in JSONL format.

7. Docker (optional)

docker build -t minivault .
docker run -p 3000:3000 -v $(pwd)/logs:/app/logs minivault

To use a custom Ollama URL (e.g., remote or non-default port):

docker run -p 3000:3000 -v $(pwd)/logs:/app/logs -e OLLAMA_URL="http://host.docker.internal:11434/api/generate" minivault

Replace http://host.docker.internal:11434/api/generate with your Ollama endpoint as needed.

---

Endpoints

POST /generate

• Input: { "prompt": "Who is Euler?" }
• Output: { "response": "..." }

GET /status

• Returns memory, uptime, and system info.

---

Design Notes & Tradeoffs

• Local-first: No cloud calls; all LLM responses are from your local Ollama instance. This ensures privacy and offline capability, but means model quality and speed depend on your local hardware and model selection.
• Logging: Uses JSONL for easy parsing and streaming. This format is append-only and simple, but could grow large over time; log rotation or archiving may be needed for production use.
• Error Handling: API returns clear error messages and logs failures. This helps with debugging and transparency, but for a production tool, more granular error codes and user-friendly messages could be added.
• Extensible: The codebase is modular, making it easy to swap out the model, add endpoints, or integrate with other local tools. However, it currently assumes a single model and endpoint; multi-model or multi-user support would require further abstraction.
• Model Selection: The model is set via environment variable. For more flexibility, the API or CLI could allow dynamic model selection per request.
• Security: The API is open by default for local use. For shared or production environments, authentication and rate limiting should be considered.