Sparkkeep – Single-binary Go/SQLite action engine for saved links and notes Sparkkeep, a self-hosted "action engine" from developer BurlacuAndrei, ships as a single ~24 MB Go binary backed by one SQLite file in WAL mode with FTS5 search, capturing links, notes, voice memos and screenshots from Telegram, Discord, Slack, a webhook or a browser extension and running them through AI triage into Now/Next/Later boards. The tool uses the user's own LLM to generate TL;DR briefs and checklists, offers deterministic research playbooks with a 5-Pillar Research Health Linter, cost estimator and zero-cost dry runs, and exports to Obsidian, Notion and Joplin; it runs in under 30 MB RAM on a Raspberry Pi, NAS or 5 € VPS. A managed Sparkkeep Cloud edition with multi-user accounts and managed AI/search infrastructure is announced as upcoming, with early access requested via GitHub issues. Capture anything. Triage it with AI. Ship what matters. Sparkkeep is a self-hosted action engine for the ideas you keep collecting and never execute. Share a link, a note, a voice memo, or a screenshot — it scrapes it, filters the noise, synthesizes an actionable brief, and puts it on a board you actually work from. One binary. One SQLite file. No Postgres, no Redis, no cloud account. You save 200 interesting links. You open maybe four. The rest rot in a folder called "later". Sparkkeep is not a bookmark manager. It is an action engine: Capture ──► Noise Gate ──► AI Triage ──► Horizon Routing ──► Execution Board ──► Research ──► Digest ──► Shipped 1. Capture from Telegram, Discord, Slack, the browser extension, a webhook, or a plain note with no URL at all. 2. Filter engagement bait and spam before it reaches your inbox. 3. Synthesize a brief — TL;DR, why you'd care, and a checklist of next actions — using your own LLM. 4. Route work into Now, Next, or Later, and execute it from a widescreen board with inline checklists. 5. Research properly: deterministic playbooks that search, cite sources, and issue a Pursue / Watch / Skip verdict instead of hallucinating. 6. Review with a 7/14/30-day digest, then push finished notes to Obsidian, Notion, or Joplin. docker run -d \ --name sparkkeep \ -p 8080:8080 \ -v sparkkeep data:/data \ ghcr.io/burlacuandrei/sparkkeep:latest Open http://localhost:8080 http://localhost:8080 . The first-run wizard asks for a passphrase and an AI provider, and offers a one-click demo library so you can look around before committing. Other install options Docker Compose services: sparkkeep: image: ghcr.io/burlacuandrei/sparkkeep:latest container name: sparkkeep restart: unless-stopped ports: - "8080:8080" volumes: - ./data:/data docker compose up -d Standalone binary — ~24 MB, Linux / macOS / Windows, from Releases https://github.com/BurlacuAndrei/sparkkeep/releases : tar -xzf sparkkeep-