Find the seam worth mining. Let Vein do the digging.
Every morning, before you open your laptop, Seam has already searched GitHub and YouTube, scored every candidate against your profile with local AI, and picked the top 3 worth studying.
$ vein morning── Today's Seam Picks (2026-06-02) ─────────────────────1.[github] simonw/files-to-prompt ★ 94/100
Why: Python CLI, solo-built, directly applicable to vein pipe
2.[github] astral-sh/ruff ★ 78/100
Why: Rust extension pattern reusable in Lode
3.[youtube] Fireship — "Tauri 2 deep dive" ★ 71/100
Why: covers wry/tao APIs you need for Lode web clipper
─────────────────────────────────────────────────────────
Run: seam fetch-today | vein study fetch daily-2026-06-02
⌨️ CLI design
seam run — full pipeline
search → score → pick → print. Runs unattended in a nightly cron.
seam run# search → score → pick 3 → printseam run --picks 5 # override daily limitseam run --source github # GitHub onlyseam run --pipe # output owner/repo lines for piping
seam run --pipe | vein fetch — seamless Vein integration
Unix pipe: Seam's picks go straight into the Vein mine.
seam run --pipe | xargs -I{} vein fetch {}
seam run --pipe | while read repo; do vein fetch "$repo" --tag daily; done
🧭 Scoring
Each candidate is scored on two weighted dimensions by local ollama (deepseek-r1:14b). No cloud LLM, no API cost, runs at 2am unattended.
Dimension
Measures
Default weight
target_relevance
How relevant to your current projects
60%
solo_feasible
Can a solo dev with your profile use / learn from this?
40%
ollama down → auto-fallback to heuristic (stars + keyword). Picks below score 60 are excluded.
🔑 Design decisions
Local AI, not cloud LLM
ollama runs on your machine. No API key management at 2am. Repo lists may be sensitive — nothing leaves your computer. Cost: zero. Graceful fallback to heuristic if ollama is down.
Official APIs only — no scraping
GitHub Search API + YouTube Data API v3. Scraping HTML is fragile; official APIs are stable and free within rate limits.
CLI pipe integration, not SDK
seam run --pipe | vein fetch. Unix philosophy. Seam survives Vein refactors. No cross-project Python dependency.
3 picks / day, not a feed
Max 3 per day, 14-day cooldown per repo. Seam selects signal, not volume.
JSON Lines, not SQLite
.seam/picks.jsonl append-only. Human-readable, grep-able, no schema migrations. SQLite is Vein's domain.
🗺 Roadmap
Phase 0 · GitHub + CLI Planned
seam run full pipeline, GitHub Search API, ollama scoring, picks log, pipe to Vein.
Phase 1 · YouTube source
YouTube Data API v3, score video titles + descriptions against profile.
Phase 2 · Multi-source
Hacker News, RSS dev blogs, cross-source unified scoring (0–100 normalised).
Phase 3 · Vein MCP integration
Seam exposes seam_picks_today MCP tool. Vein MCP calls it directly — no manual pipe.
🛰️ Seam — In Planning
The geological survey bot is being built. Phase 0 (GitHub CLI) in development. Drop a note to follow progress.
