Lookout

A personal search engine that filters web results through your daemon profile, plus a discovery mode that finds values-aligned humans on GitHub.

Google personalizes search based on surveillance. Lookout personalizes based on declaration.

What it does

/search — your search query hits the Brave Search API for raw web results, then Claude re-ranks them against your daemon profile. You get back 5 results scored 1-10 with a one-line note on why each one matters to you, specifically.

/discover — adjacency-mining mode for finding values-aligned humans. Pulls contributors from N anchor repos (defaults: substrate-first projects like loam, mycelia, modelcontextprotocol/servers, bluesky-social/atproto), builds a candidate dossier for each (profile + top non-fork repos + topics), then Claude re-ranks against your values profile. Returns ranked candidates with values_score, the signals that landed, a suggested intro paragraph in your voice, and reachable contact paths.

The whole thing runs on a single Cloudflare Worker. No database, no state, no tracking.

What's a daemon profile?

The concept comes from Daniel Miessler's daemon / personal API — a structured declaration of who you are, what you know, and how you want information delivered. Your daemon profile lives in src/index.ts as a template string (DAEMON_PROFILE). Edit it to match you.

For /discover, there's a parallel VALUES_PROFILE constant that captures what you're looking for in other people. Tune it the same way.

Quick start

# Fork this repo
git clone https://github.com/YOUR_USERNAME/lookout.git
cd lookout

# Install dependencies (bun preferred, npm works)
bun install

# Edit your daemon profile + values profile in src/index.ts
# Replace the template fields with your actual identity

# Set your secrets
wrangler secret put BRAVE_SEARCH_KEY     # Brave Search API token
wrangler secret put ANTHROPIC_API_KEY    # Claude API key
wrangler secret put LOOKOUT_API_KEY      # bearer token YOU pick (protects all non-/health endpoints)
wrangler secret put GITHUB_TOKEN         # GitHub PAT, read-only scope (for /discover rate limit headroom)

# Deploy
wrangler deploy

For local development:

# Create .dev.vars with your keys (see .dev.vars.example)
cp .dev.vars.example .dev.vars
# Edit .dev.vars with real keys

# Run locally
bun run dev

Auth

Every endpoint except /health requires a bearer token. Pass LOOKOUT_API_KEY via either:

• HTTP header: Authorization: Bearer YOUR_KEY
• Query param: ?key=YOUR_KEY (for the HTML UI, which also stores it in sessionStorage)

Unauthorized requests return 401.

Rate limits

• /search — 10 requests/minute/IP
• /discover — 2 requests/minute/IP (each call makes ~150-200 GitHub API requests + one big Claude call)

Security

• HSTS, CSP, X-Content-Type-Options, X-Frame-Options, Referrer-Policy, Permissions-Policy applied to every response
• Repo slug and GitHub login validated against strict regex (anti path-injection)
• Anchor count capped at 10, candidate count capped at 60, output capped at 25
• Error responses are sanitized — no leakage of GitHub or Claude API error details
• Query length capped at 500 characters

Cost

Cheap enough to run personally:

• Brave Search: Free tier 2,000 queries/month; paid starts at $5/mo for 20K
• Anthropic (Claude Sonnet): ~$0.003 per /search call, ~$0.10 per /discover call
• GitHub API: Free with a PAT (5,000 req/hr authenticated rate limit; /discover consumes ~150-200 per call)
• Cloudflare Workers: Free tier 100K requests/day

At typical personal use (10-30 searches/day, occasional discover runs), under $5/month total.

API

POST /search

curl -X POST https://your-worker.workers.dev/search \
  -H "Authorization: Bearer YOUR_LOOKOUT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"query": "rust async runtime internals", "count": 5}'

Response:

{
  "query": "rust async runtime internals",
  "results": [
    {
      "title": "...",
      "url": "...",
      "snippet": "...",
      "daemon_score": 9,
      "daemon_note": "Deep implementation walkthrough with code — exactly your level"
    }
  ],
  "daemon": "lookout",
  "ts": "2026-06-13T..."
}

POST /discover

curl -X POST https://your-worker.workers.dev/discover \
  -H "Authorization: Bearer YOUR_LOOKOUT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "mode": "adjacency",
    "anchors": ["NorthwoodsSentinel/loam", "modelcontextprotocol/servers"],
    "count": 15,
    "per_anchor": 15
  }'

anchors is optional. If omitted, uses the defaults baked into src/index.ts. count capped at 25, per_anchor between 5 and 30.

Response:

{
  "mode": "adjacency",
  "total_candidates": 42,
  "results": [
    {
      "login": "...",
      "name": "...",
      "bio": "...",
      "html_url": "https://github.com/...",
      "values_score": 9,
      "values_notes": "Three sentences explaining the specific signals.",
      "suggested_intro": "A paragraph you could send to start the conversation.",
      "reach_via": ["github", "blog:https://...", "email:..."],
      "contributed_to": ["NorthwoodsSentinel/loam", "NorthwoodsSentinel/mycelia"]
    }
  ],
  "daemon": "lookout-discover",
  "ts": "2026-06-13T..."
}

GET /health

No auth required. Returns { status: "ok", daemon: "lookout", version: "0.2", features: ["search", "discover"], ts: "..." }.

Project structure

src/index.ts     # Everything — daemon profile, values profile, search, discover, UI, auth, rate limits
wrangler.toml    # Cloudflare Worker config

Yes, it's one file. That's the point.

Versions

• v0.2 (June 2026) — added /discover for values-aligned human discovery on GitHub
• v0.1 (March 2026) — initial release: identity-lensed search via Brave + Claude

---

Northwoods Sentinel Labs

Part of the Northwoods Sentinel Labs ecosystem — open-source tools for human-centered AI.

Blog · Substack · GitHub