The open-core engine watches repeated local file workflows and suggests automations you approve. Private Intelligence decides which suggestions surface, when, and how they’re worded — so the CLI stays quiet and useful. Still on your machine.
Open-core engine · Private Intelligence for decision quality · local SQLite · allowlist · dry-run · undo
The engine can be open. Decision quality is the product — ranking, timing, suppression, and wording that keep suggestions worth reading.
Capture, patterns, baseline suggestions, approve, dry-run, run, undo, and path allowlist. MIT, inspectable, useful alone.
Decides which suggestions surface, when they appear, and how they’re labeled — from your local feedback history.
Private ≠ cloud. Intelligence evaluates locally. Turn it off and open-core still works. That boundary is the business model: infrastructure open, decision quality as the product.
With intelligence on, flowctl suggestions shows fewer low-value items, better order, and stable reason codes via suggestions --explain (e.g. timing.too_new, suppression.cooldown_after_reject).
It does not invent filesystem facts, skip approval, run automations, or ship your capture data to a remote brain.
Build with --features intelligence, set intelligence_enabled = true, and install a local license at ~/.config/flowd/intelligence.license.toml (or FLOWD_INTELLIGENCE_DEV=1 for development). Docs: Private Intelligence.
Open-core stays free. Private Intelligence is the product layer — still local, unlocked with a license file.
Observe, baseline suggestions, approve → dry-run → run → undo, path allowlist, audit trail.
Try Private Intelligence on your machine with an evaluation license.
Quieter CLI for daily use — decision quality as the product.
Pro plus shared policy packs and a commercial support lane.
Prices: contact for current Eval / Pro / Team terms — we keep numbers out of the OSS tree until tiers stabilize. License check is local (`flowctl doctor`).
No blank canvas. No Zapier-style wiring. You keep working; open-core notices what repeats. Intelligence decides what is worth showing.
Filesystem events in folders you choose. Optional clipboard, downloads, and terminal bridges.
Open-core turns rename/move sequences into patterns with counts and baseline scores.
Private Intelligence ranks, delays, or suppresses — so the CLI stays quiet.
Dry-run first, then run. Undo and runs show keep every apply inspectable.
Hardening on the road to 1.0 — engine trust first, intelligence as the quality layer.
Live client wiring, anti-annoyance caps, delay freshness, preference memory, and optional localhost LLM labeling for explain.
Dry-run, run, undo, and daemon auto-run stay inside observed_folders / execution_allowed_roots, including symlink escape checks.
flowctl daemon *, crash-recovery soak tests, and flowctl runs show for every dry-run / apply / undo.
Workflow pack registry, terminal mv/cp into directories, team policy packs (can force intelligence off).
You don’t have to choose between automation and keeping work on your machine.
Events stay in a local SQLite database. No accounts, no sync of capture data, no telemetry.
Only folders and optional bridges you enable. Clipboard and terminal observation stay off by default.
flowctl status, suggestions --explain, runs show, and plain SQL. See privacy docs.
Most tools wait for you to describe the workflow. flowd starts by watching — from the terminal.
v0.3 installs from source (or Homebrew HEAD). You need a stable Rust toolchain.
# clone & install git clone https://github.com/nickvd7/flowd && cd flowd ./scripts/install.sh # or: cargo install --path crates/flow-cli && cargo install --path crates/flow-daemon # first-run setup + daemon flowctl setup --watch ~/Downloads --watch ~/Desktop flowctl daemon start flowctl daemon status # core loop (open-core) flowctl suggestions flowctl approve 1 flowctl dry-run 1 flowctl run 1 flowctl runs show 2 # Private Intelligence (sibling flowd-intelligence checkout) cargo install --path crates/flow-cli --features intelligence --force # config: intelligence_enabled = true # license: ~/.config/flowd/intelligence.license.toml (or FLOWD_INTELLIGENCE_DEV=1) flowctl doctor flowctl suggestions --explain # optional Homebrew HEAD (open-core) brew install --HEAD --formula Formula/flowd.rb
Short answers before you install.
The product layer on top of open-core. It ranks, times, suppresses, and phrases suggestions so the CLI stays quiet and useful. Open-core owns facts and actions; intelligence only influences presentation. See docs/intelligence.md.
No and no. Open-core works alone with baseline suggestions. Private means a closed decision policy evaluated locally — not a cloud brain. Capture stays in SQLite on your machine.
No. Observation, patterns, suggestions, and automations stay in local SQLite. Open-core has no cloud sync of capture data. Optional local LLM labeling talks only to localhost.
Filesystem events in folders you configure. Optional clipboard metadata, browser-download bridge records, and terminal-history bridge lines — all off by default except folder watching.
Not when the allowlist is enforced (default). Destinations must stay under observed_folders or explicit execution_allowed_roots. Symlink escapes are rejected.
After a rename/move workflow repeats enough times in watched folders. For rare workflows, use flowctl teach from-session.
No. Approval is required. A dry-run is required before flowctl run unless you pass --force. Daemon auto-run stays off by default.
Not for local files. Bridges are optional NDJSON files for downloads/terminal history — no SaaS connectors.
Two lanes: public open-core issues on GitHub, private product questions here.
Bugs, setup help, suggestion-quality / dogfood notes, and features — via templates so we can triage.
Evaluation access, Private Intelligence questions, partner / paid support — use the form below (not public issues).
Private line for Intelligence / commercial support. For open-core bugs, prefer GitHub Issues.