prism
Health Warn
- License — License: MIT
- Description — Repository has a description
- Active repo — Last push 0 days ago
- Low visibility — Only 5 GitHub stars
Code Warn
- network request — Outbound network request in web/app.js
Permissions Pass
- Permissions — No dangerous permissions requested
No AI report is available for this listing yet.
Programmable Responsive Interface for Smart Models
PRISM
Programmable Responsive Interface for Smart Models
Also known as: Probably Runs Interesting Stuff Magically
Yes, another AI dashboard. Except this one runs entirely on your own hardware — no cloud, no API keys, no data leaving your machine — and has the slightly unsettling property of being able to modify its own environment.
More screenshots
The idea
This started as frustration with tools like OpenClaw and Hermes — both promising, both Node.js (make of that what you will), and both operating on the same fundamental assumption: the agent lives somewhere in the background, has direct access to your disk, and you talk to it through a chat box that it has absolutely no influence over. Which is fine, until you realize that a truly useful agent should be able to shape its own workspace, not just occupy it.
PRISM is built around a different premise: the agent has actual agency over its environment. It runs code in a sandboxed workspace (not on your host machine — you're welcome), builds interactive widgets and pins them directly to the dashboard it lives in, schedules tasks, searches the web, queries a private knowledge base, and — if it needs a capability it doesn't have — it can define new tools and use them immediately, or connect a remote MCP server and configure it entirely on its own.
You ask it to monitor something, it builds a widget. You ask it to set up a GitHub integration, it connects the MCP server, fetches its auth token, and gets to work. You ask it to remember a document, it indexes it with embeddings and will bring it up when relevant.
Is this a great idea? Probably. Does it make you slightly nervous? It should. That's how you know it's doing something real.
Stack
| Layer | Tech |
|---|---|
| Backend | Go |
| Frontend | Vanilla JS, GridStack |
| LLM | Ollama (any model) |
| Embeddings | Ollama (any embedding model) |
| Vector store | PostgreSQL + pgvector |
| Web search | SearXNG |
| Runtime | Docker / Docker Compose |
Requirements
- Docker + Docker Compose
- An Ollama instance (local or remote) with at least one model and one embedding model pulled
Quick start
git clone https://github.com/blazux/prism
cd prism
cp .env.example .env # edit to point at your Ollama instance
docker compose up -d
Open http://localhost:48080.
Configuration
All configuration is done via environment variables (set in docker-compose.yml or a .env file):
| Variable | Description | Example |
|---|---|---|
OLLAMA_URL |
URL of your Ollama instance | http://ollama:11434 |
OLLAMA_MODEL |
LLM model to use | qwen3.6:27b |
EMBED_MODEL |
Embedding model for RAG | qwen3-embedding:8b |
POSTGRES_URL |
PostgreSQL connection string | postgres://rag:rag@postgres:5432/rag |
SEARXNG_URL |
SearXNG instance URL (optional) | http://searxng:8080 |
WORKSPACE_DIR |
Agent workspace directory | /workspace |
PLUGIN_DIR |
Widget storage directory | /workspace/.plugins |
TZ |
Timezone for the agent and cron jobs | America/New_York, Europe/Paris |
The only required changes are
OLLAMA_URLand the model names — everything else works out of the box with the default Docker Compose setup.
Note on embedding models: the vector dimension is detected automatically at startup by probing the model. If you change the embedding model, you need to reset the RAG database (the vector dimension is fixed per table).
Note on timezone:
TZdefaults toUTC. Set it to your local timezone (IANA format, e.g.America/New_York,Europe/Paris,Asia/Tokyo) so that cron schedules and agent timestamps match your local time. For a fixed offset without daylight saving, useEtc/GMT+4(note: POSIX convention inverts the sign —Etc/GMT+4= UTC-4).
License
MIT
Reviews (0)
Sign in to leave a review.
Leave a reviewNo results found