Skip to content
Projects

boffti/stardash has 34 GitHub stars and 10 forks. GitHot measured 0 stars in 24 hours. Last checked 2026-11-15 14:42 UTC. Last push 2026-11-10.

How these counts are measured

What you can do

  • RadarA nightly inbox of new and rising repos from GitHub search, Hacker News and the people you follow, ranked against your stars, each with its reasons. - All StarsUp to 5,000 stars, searchable across names, descriptions, notes, tags and collections, with a status, notes and a pin on each.
  • AI categorizeOne pass proposes collections and tags from your stars and files every repo. It reuses your taxonomy on later runs. - MCP serverSeven tools for any MCP client, on bearer tokens you mint and revoke in Settings or from the CLI.
  • Repo IntelA health brief per repo: score, maintenance verdict, community signals, pain points and a recommendation. - ContributeGood first issues from repos you already use, with a brief and a ready-to-paste prompt for your coding agent.
  • DiscoverDescribe what you are building. AI expands the query and reranks GitHub search against your stars. - One containerOne image, one SQLite file, one volume. amd64 and arm64. Nightly jobs run in-process. No Postgres, no Redis.

Measured star growth

in 24h
0
in 7 days
Collecting history
in 15 days
Collecting history
in 1 month
Collecting history

Net star changes over overlapping periods, not daily totals. Missing measurements are left blank. Stars measure attention, not software quality.

About

AI-powered workspace to discover, organize, and contribute to your starred GitHub repos - with MCP support for code assistants (coming soon)

Developer Tools

README.md

Original
StarDash. Your GitHub stars, organized. New repos your taste says to look at.

You star repos and forget them. StarDash remembers every one, sorts them for you, and every night finds the new repos your own stars say you will want.

CI status License: MIT Docker image on GHCR Works with Claude Code, Cursor and any MCP client Runs on your machine

Why StarDash · Quick start · How to use it · A look inside · Features · Roadmap · Contributing · 简体中文

StarDash at full window. Radar opens on an inbox of new repos ranked against the user's stars, each row saying why it surfaced. The user moves down with j, dismisses with x and undoes it, filters to people they follow and to Hacker News, opens the From your stars tab, then All Stars, a collection, a repo page, Intel health reports, Contribute issues and the MCP settings.

Why StarDash

Starring is how developers bookmark. Then the list grows past a few hundred and stops being useful. StarDash fixes both halves of that:

  • Find what you starred. Every star is searchable by name, description, your notes, tags and collections. Mark each one want-to-try, currently-using, tried-liked, tried-dropped, just-interesting or reference, so the list says what you did with it.
  • Sort it without the chore. One AI pass reads up to 500 of your stars and proposes 5 to 12 collections and 15 to 25 tags, then files everything. Keep, rename or delete what it made.
  • Hear about the next one. Radar is an inbox of new and rising repos, ranked by how well they fit what you already star, and every row says why it is there.

Ranked by your taste, not by GitHub's. Radar builds a profile from your stars: the topics and languages you keep coming back to, weighted by what you did with each repo. A tried-dropped star counts against its topics. Every night it pulls candidates from GitHub search, Hacker News and the people you follow, and scores each one against that profile. A star from someone you follow outranks a topic match.

Every row explains itself. No black-box score. A row says Matches ai, llm, python in your stars, 3 people you follow starred this, +2.1k stars this week or 40 points on Hacker News. Snooze it for a week or dismiss it from the keyboard, or star it in one click.

Your agent can use it. StarDash ships an MCP server. Claude Code, Cursor or any MCP client can search your stars, file them, and triage Radar on the subscription you already pay for.

Yours, on your machine. One container, one SQLite file. Sign-in creates a GitHub App that you own, so no one else's name is on the consent screen and no token leaves your box.

Same feed, three people

Radar reads the same pool of new repos for everyone. What reaches the top depends on the stars behind it:

You mostly star What Radar puts first, and the reason it shows
Rust CLIs and terminal tools A new Rust IDE with 40 points on Hacker News. Matches developer-tools, rust, terminal in your stars.
LLM agents and MCP servers An agent memory store created this week. Matches ai, llm, python in your stars. Python, your usual.
Self-hosted apps you run at home A home-lab dashboard three people you follow starred yesterday. People you follow outranks every topic match.

[!NOTE] AI is optional. Search, status, notes, Radar and browsing the Contribute issue list work without a key. Generating a Contribute brief or coding-assistant prompt requires a configured AI provider. Categorize, Intel and Discover also require AI. Configure an OpenRouter key (OPENROUTER_API_KEY), paste your own OpenAI or Anthropic key in Settings, or use a local model through an OpenAI-compatible server such as Ollama (AI_BASE_URL). A full categorize run over 500 repos costs about a cent on the default model.

Quick start

You need Docker.

docker run -d --name stardash -p 3000:3000 -v stardash-data:/data ghcr.io/boffti/stardash:latest

Open http://localhost:3000 and press Connect GitHub. GitHub shows a page that creates a GitHub App owned by you: press Create GitHub App, then Authorize. There is no OAuth app to register and no secret to copy. Your stars sync, and Radar fills on the first nightly scan, or right away when you press Scan now.

Alternative: install with Docker Compose

Compose is an alternative fresh install; use it instead of the docker run command above, not alongside it. It publishes the same host port (3000) and creates its own named volume (normally stardash_stardash-data), so it starts a separate instance with a separate database.

curl -O https://raw.githubusercontent.com/boffti/stardash/main/docker-compose.yml
OPENROUTER_API_KEY=sk-or-... docker compose up -d

To update either installation, pull the image and recreate its container. Keep using the same data volume.

Moving an existing docker run install to Compose: stop and remove the old container without deleting its volume, then edit the downloaded Compose file's top-level volume declaration to reuse that existing volume:

volumes:
  stardash-data:
    external: true
    name: stardash-data

Then start Compose. Do not run both containers on port 3000, and do not omit the external mapping when you intend to keep using the existing database.

How it works

flowchart LR
    S["Your stars<br/>and statuses"]:::green --> P["Taste profile<br/>topics, languages"]:::steel
    G["GitHub search"]:::clay --> C["Nightly candidates"]:::steel
    H["Hacker News"]:::clay --> C
    F["People you follow"]:::clay --> C
    P --> R["Radar: scored,<br/>with reasons"]:::green
    C --> R
    R --> T["You star, snooze<br/>or dismiss"]:::steel
    T --> S
    classDef clay fill:#b8572f,stroke:#9c4526,color:#ffffff
    classDef steel fill:#2f6497,stroke:#244c77,color:#ffffff
    classDef green fill:#008a1c,stroke:#006e16,color:#ffffff

Next.js 16 in one process, SQLite through node:sqlite, no external services. Two jobs run in-process every night: star snapshots, which give Radar its velocity numbers, and the Radar ingest from GitHub search, Hacker News (through Algolia) and the stars of people you follow. The whole scoring function is 156 lines in lib/radar.ts.

How to use it

Triage Radar from the keyboard

Key What it does
j / k Move down and up the inbox
x Dismiss. It stays gone. A toast offers Undo.
z Snooze for a week
o Open on GitHub

The tiles above the list filter it: People you follow, Hacker News, Rising, New this week. From your stars shows recommendations mined from your own recent stars.

Ask your agent

In Settings, open Agents (MCP), press New token, and paste the config it shows into .mcp.json (Claude Code) or your client's MCP settings:

{
  "mcpServers": {
    "stardash": {
      "type": "http",
      "url": "http://localhost:3000/api/mcp",
      "headers": { "Authorization": "Bearer sd_…" }
    }
  }
}

Then talk to your agent in plain words:

You say What happens
"What did I star about vector databases?" search_stars looks across names, descriptions, your notes, tags and collections.
"Tell me about duckdb and what I wrote about it." get_repo returns the repo with your status, notes, tags and collections.
"File every MCP server I starred under Agents and tag them mcp." list_taxonomy and organize_repo create and assign tags and collections. Sync your GitHub stars in StarDash first; organize_repo only works on repos already synced into your instance.
"What's on my radar this week?" radar_feed returns the ranked inbox with each row's reasons.
"Dismiss anything on my radar that is another Electron wrapper." radar_action dismisses, snoozes or undoes.
"Star the top two and put them in Want to try." star_repo stars them on GitHub. Sync your stars in StarDash, then use organize_repo to assign a status, tags or collections.

A look inside

From the inbox you check over coffee to the repo you end up using.

Radar

Radar inbox: stat tiles for each signal, then ranked rows with the repo, the reasons it matched and a signal badge.
Inbox. New and rising repos ranked against your stars. Each row says why it matched, and the badge names its strongest signal.
Radar filtered to repos starred by people the user follows.
People you follow. One tile per signal filters the inbox. A star from someone you follow ranks above any topic match.
From your stars: card grids of recommendations grouped by theme.
From your stars. Recommendations mined from your own recent stars, grouped by what they have in common.
Settings, Agents section: mint a token and copy the MCP config.
Agents (MCP). Mint a token, copy the config, and your agent can search, file and triage.

Your stars

All Stars: a grid of starred repos with tags, language, stars and forks.
All Stars. Every star, searchable by name, description, notes, tags and collections.
A collection view: the repos the AI filed under AI Agents and Skills.
Collections. Yours, or proposed by one AI pass over your stars and filed for you.
The obra/superpowers repo page: header with stars and forks, topics, ready-to-work issues, the README and an Intel health score of 81.
One repo. The README inline, its topics, a health card, and open issues ready for a first contribution.
Repo Intel: health score rings and briefs per repo.
Intel. A health brief per repo: score, maintenance verdict, community signals and pain points.
Contribute: open issues from starred repos ranked by fit, with difficulty and type filters.

Contribute. Open issues from repos you already starred, ranked by fit, with a one-click brief and a prompt for your coding agent.

Features

Radar
A nightly inbox of new and rising repos from GitHub search, Hacker News and the people you follow, ranked against your stars, each with its reasons.
All Stars
Up to 5,000 stars, searchable across names, descriptions, notes, tags and collections, with a status, notes and a pin on each.
AI categorize
One pass proposes collections and tags from your stars and files every repo. It reuses your taxonomy on later runs.
MCP server
Seven tools for any MCP client, on bearer tokens you mint and revoke in Settings or from the CLI.
Repo Intel
A health brief per repo: score, maintenance verdict, community signals, pain points and a recommendation.
Contribute
Good first issues from repos you already use, with a brief and a ready-to-paste prompt for your coding agent.
Discover
Describe what you are building. AI expands the query and reranks GitHub search against your stars.
One container
One image, one SQLite file, one volume. amd64 and arm64. Nightly jobs run in-process. No Postgres, no Redis.

Your data

[!IMPORTANT] Your stars, statuses, notes, taxonomy, Radar history and GitHub App credentials live in one SQLite file on the /data volume. StarDash communicates with GitHub and Hacker News's search API, and sends AI requests to your configured AI provider when you use AI features. Optional observability integrations also send telemetry to their configured services: Sentry receives error/performance events when a Sentry DSN is set; Langfuse receives AI traces only when both LANGFUSE_PUBLIC_KEY and LANGFUSE_SECRET_KEY are set.

What Where
Stars, notes, statuses, tags, collections stardash.db on the volume
GitHub App id and secret stardash.db, created on your account during setup
Your sign-in A signed cookie whose secret is minted on first start
MCP tokens Only their SHA-256 hashes are stored

Single owner: the first GitHub account to sign in owns the instance.

Configuration

All optional.

Variable Purpose
OPENROUTER_API_KEY Turns on the AI features.
AI_MODEL Any OpenRouter model id. The default is a cheap model that handles structured output.
AI_BASE_URL An OpenAI-compatible endpoint (Ollama, LM Studio, vLLM). Used instead of OpenRouter; needs AI_MODEL.
AI_API_KEY Key for AI_BASE_URL, if the server wants one.
APP_URL Public origin, when a reverse proxy does not send X-Forwarded-Host and X-Forwarded-Proto.
GITHUB_CLIENT_ID / GITHUB_CLIENT_SECRET Use a GitHub App you made yourself instead of the one-click flow.
GITHUB_PAT Token for the nightly jobs. Defaults to the owner's.
CRON_SECRET Lets an external scheduler call /api/cron/*. DISABLE_SCHEDULER=1 turns the in-process jobs off.
NEXT_PUBLIC_RADAR_ENABLED false hides Radar.
DATABASE_PATH The SQLite file. /data/stardash.db in the container.

Local models: run Ollama on the host, pull a model that handles JSON output well (ollama pull qwen3:8b), then set AI_BASE_URL=http://host.docker.internal:11434/v1 and AI_MODEL=qwen3:8b. From pnpm dev use http://localhost:11434/v1. An 8B model needs roughly 8 GB of free RAM. Models under 2B run but group poorly (everything lands in one collection); if Categorize or Intel fails or looks wrong, try a larger one.

No browser on the box: choose Paste a personal access token instead on the login page. StarDash talks to GitHub.com only; GitHub Enterprise Server is not supported yet.

Plain HTTP on a LAN works. The session cookie is not marked Secure there. Put it behind HTTPS for anything public.

Reverse proxy

Terminate TLS at the proxy and forward the original host and protocol. The GitHub App setup and OAuth callback URLs are built from that origin.

Caddy

stardash.example.com {
    reverse_proxy 127.0.0.1:3000 {
        header_up Host {host}
        header_up X-Forwarded-Host {host}
        header_up X-Forwarded-Proto {scheme}
    }
}

nginx

server {
    listen 443 ssl;
    server_name stardash.example.com;

    # ssl_certificate and ssl_certificate_key go here.
    location / {
        proxy_pass http://127.0.0.1:3000;
        proxy_set_header Host $host;
        proxy_set_header X-Forwarded-Host $host;
        proxy_set_header X-Forwarded-Proto $scheme;
    }
}

Keep port 3000 private and let the proxy take public traffic. APP_URL wins when set; otherwise the origin comes from X-Forwarded-Host and X-Forwarded-Proto, then the request URL. Set APP_URL=https://stardash.example.com (no path, no trailing slash) when your proxy cannot pass those headers reliably or you want to pin the canonical origin. Only trust forwarded headers from a proxy you control.

Without Docker: Node 22.5 or later and pnpm, then pnpm install && pnpm build && pnpm start.

Roadmap

  • One container on SQLite, with sign-in through a GitHub App you own
  • Radar: nightly candidates from GitHub search, Hacker News and people you follow, ranked against your stars
  • MCP server and CLI tokens
  • AI categorize, Intel, Contribute and Discover on cheap models
  • v1.0.0 release
  • Drive the AI features from your agent's subscription through the CLI, with the API key as a fallback
  • Weekly releases, each with one visible change

ROADMAP.md has the full plan. Have an idea? Open an issue.

Contributing

pnpm install && pnpm dev     # sign in at http://localhost:3000; you are running the real thing
pnpm typecheck && pnpm lint && pnpm check:db && pnpm check:radar

Issues labelled good first issue are small and self-contained. CONTRIBUTING.md has the conventions. To report a security problem, follow SECURITY.md.

For AI agents

Start with AGENTS.md. It covers the schema in lib/db.ts, the auth flow, Radar scoring, the MCP tools in lib/mcp-tools.ts, and the UI rules every page follows.

Command line reference

bin/stardash.mjs works directly on the SQLite file. In the container, prefix it with docker exec stardash node.

bin/stardash.mjs token [name]       # mint an MCP token and print it once
bin/stardash.mjs tokens             # list tokens and when each was last used
bin/stardash.mjs revoke <id>        # revoke a token
bin/stardash.mjs mcp-config [tok]   # print the .mcp.json block for a token

License

MIT