Identity-lensed search — your daemon is the filter. Open source personal search engine on Cloudflare Workers.
src/index.ts file with no database or persistent state.| Rating | Summary | |
|---|---|---|
| Quality | decent (14/24) | Solid concept with a clean README and minimal footprint, but no release, no tests, and negligible adoption signal. |
| PAI Relevance | watch (0.63) | Fills a real gap in PAI's search personalization layer and is TypeScript/Cloudflare-native, but overlaps partially with the Research skill and is too early-stage to integrate directly. |
14/24 — stale-risk / adequately-documented / early-or-minimal
Failed:
latest_release is none..github/workflows/.Passed:
archived: false confirmed.Failed:
/docs directory.Passed:
git clone, npm install, wrangler secret put, and wrangler deploy steps, plus a .dev.vars local dev flow.## API section contains a complete curl code block with request and response JSON.## API heading with documented endpoint, payload, and response schema including daemon_score and daemon_note fields.Failed:
package.json; README contains no mention of tests or test tooling; no CI config referenced.Passed:
package.json present and provided in dependency manifest.devDependencies (@cloudflare/workers-types, typescript, wrangler); no runtime dependencies at all.| Dimension | Score | Assessment |
|---|---|---|
| Harvest Value | 1 | The daemon-profile-as-reranking-lens pattern is worth studying: it formalizes a structured self-declaration as a prompt prefix that reshapes external API output rather than building a separate retrieval layer. PAI's Research skill does web investigation but doesn't apply an identity-shaped filter to result scoring — the scoring schema (daemon_score, daemon_note) is a concrete design worth adapting. |
| Integration Readiness | 2 | TypeScript, Cloudflare Workers, no runtime dependencies, structured JSON output from a POST endpoint — perfectly aligned with PAI's TypeScript/Bun/Cloudflare stack. Can be forked directly, the daemon profile edited as a template string, and called from a PAI skill via fetch. |
| Overlap Risk | 1 | Partial overlap with PAI's Research skill, which performs web investigation and synthesis. However, Research doesn't currently expose a daemon-lensed re-ranking path or a per-result scored/annotated output format; the overlap is functional-category rather than implementation-level. |
| Gap Fill | 1 | PAI has Research for web investigation but no dedicated personalized-search layer that filters external results through the user's identity declaration. This addresses that gap specifically, though it is narrow in scope and the gap is not critical to PAI's current operation. |
Composite: 0.63
Before investing in identity-filtered search as a pattern: Fork the repo, write a 10-line daemon profile for one narrow domain (e.g., secondhand bookselling in France, or TypeScript tooling), deploy to a Cloudflare Workers free tier with personal Brave and Anthropic keys, and run 20 test queries against a known information need. This validates whether structured self-declaration measurably outperforms raw Brave results in that domain — or surfaces where the LLM ranking step breaks down — before committing to extending the approach.
Monitor whether a stable daemon-profile schema emerges: The daemon profile as a portable, structured identity seed (name/skills/interests/anti-preferences) is the repo's most transferable idea. Watch whether other builders in the PAI space converge on a common schema or interchange format for this kind of self-declaration — if a de-facto standard appears, it becomes worth adopting as a shared context artifact across multiple tools rather than inventing a local format.
Revisit when the single-file architecture breaks apart: The entire implementation is one src/index.ts with no tests, no CI, and no version history beyond a prototype. Re-evaluate when (if) the project separates the daemon-profile ingestion, the Brave fetch, and the ranking prompt into distinct, testable units — at that point the identity-filter layer becomes independently extractable and the codebase becomes worth reading as a design reference.
Category: Personal AI & Knowledge
In this category: tinyhumansai--openhuman (excellent, skip)
Standing: Where openhuman targets broad personal AI superintelligence, lookout addresses a narrower slice — search personalization through identity declaration — making it complementary rather than competitive within the category, though both lack meaningful adoption.
Density: 7/10 — README (full, substantive), dependency manifest (complete), repo metadata (stars, forks, dates, license, language), description. Missing: source code beyond what README describes, test files, CI configuration, any tagged release, changelog, or issue history.
This is the fourth NorthwoodsSentinel repo appraised (alongside loam, brook, meridian-protocol, pii-scrub). The author is building a coherent ecosystem of lightweight Cloudflare Worker tools around the daemon/personal-API concept. Lookout is the most concrete and immediately deployable of the group — the concept is clear, the implementation is self-contained, and the cost model is honest. The "single file, no state" constraint is a deliberate design choice rather than a limitation. The daemon profile concept as a reranking input is worth tracking as a pattern even if the repo itself never gains significant adoption.