signalpipe-daemon

The user-side sender for SignalPipe v4.

SignalPipe is the judgement layer for outreach. We never see your Reddit or X credentials and never see your LLM keys. The math runs on us. The sending runs on you.

This daemon holds a Server-Sent-Events stream open to your SignalPipe brain, receives missions the brain has already scored, drafted, and approved, posts them with your own platform credentials, and acknowledges the result. It never sees your LLM keys, never scores or drafts anything itself, and keeps no copy of your pipeline on this machine. It sends, and it can read feeds on this machine for the brain to judge (read, preview). (Your leads and prospects live in the managed brain, scoped to your account.)

Install

pip install "signalpipe-daemon[all]"     # both senders + the feed reader
pip install "signalpipe-daemon[reddit]"  # Reddit only
pip install "signalpipe-daemon[twitter]" # X / Twitter only
pip install "signalpipe-daemon[reader]"  # read and preview feeds on this machine
pip install signalpipe-daemon            # base; pulls only `requests`

The platform SDKs (praw, tweepy) are optional extras and are imported only when a mission for that channel actually arrives — install just the ones you use.

Configure

Copy .env.example to .env and fill it in, or export the variables directly. A real environment variable always wins over a .env entry.

Your platform credentials stay on your machine. They are used only to talk to Reddit / X directly and are never sent to SignalPipe.

Reddit API access. Since 11 November 2025 Reddit issues new API apps only after a manual approval (its Responsible Builder Policy). If you do not already have a script app, leave the REDDIT_* variables unset: approve drafts in your agent, post them yourself, and mark them sent. X sending needs an X API account on pay-per-use pricing (since February 2026), billed to you by X.

Run

signalpipe-daemon status          # validate the key, print queue depth, exit
signalpipe-daemon run             # stream missions and send them
signalpipe-daemon run --dry-run   # log intended sends without posting or acking

run holds the stream open and reconnects with exponential backoff on any network drop. A rejected key (401) is fatal and exits non-zero; everything else is treated as transient. Stop with Ctrl-C.

Read your feeds on this machine

Some of your stations can be read from your own machine instead of by the brain. The brain marks them read_by: "client" in /stations/list; the reader fetches those feeds from here, at most 50 posts per feed with a pause between feeds, and sends each page to the brain for judging. The posts then go through the same scoring, judges and missions as any other station, and approved missions reach run as usual.

signalpipe-daemon read            # a pass every 30 minutes, until Ctrl-C
signalpipe-daemon read --once     # one pass and exit, for cron

Run it beside run, in its own terminal or service. A station the brain has just judged is skipped for a few minutes, so an extra pass costs nothing.

Check a feed before you add it

Which communities you listen to decides results more than any setting. preview reads a candidate feed on this machine, sends its posts to the brain, and the brain's judges read the ones that match your product best. Nothing is saved.

signalpipe-daemon preview --product <product-id> --url https://www.reddit.com/r/SUBREDDIT/new/.rss

The verdict comes from thresholds fixed in advance: VIABLE (real buyers post here), MARGINAL (a trickle), ON-TOPIC, NOT IN-MARKET (people discuss your topic but nobody is asking to buy), NO BUYERS, or NO DATA (nothing could be judged; the explanation says why, such as an empty or stale feed or keywords that blocked every post). Each judged post costs one judgement (8 by default, --sample changes it). If this machine cannot read the feed, the brain is asked to fetch it.

How a mission flows

1. The brain scores a signal, drafts a reply, and (once approved) marks the mission ready. It streams the mission to this daemon over SSE.
2. The daemon posts the pre-written draft on the mission's channel (twitter_reply, reddit_comment, or reddit_dm) using your credentials. Missions on the manual channel are skipped — you send those yourself. A reddit_dm goes only to someone who asked for a private message: Reddit and X both require the recipient's consent before an app sends one. X also bans automated replies to posts found by keyword search, and Reddit acts against automated comments, so twitter_reply, reddit_comment and reddit_dm missions are never auto-approved; each waits for you. The drafts are written by AI: Reddit asks that AI-generated content be disclosed, so read and edit each one in your own words before you approve it.
3. The daemon acks the outcome. A failure tagged banned or rate_limited tells the brain to pause your stream for a cool-down.

Daily caps pace your sending; when a cap is hit the mission is skipped, not failed, so it remains queued and retries after the next local-midnight reset.

Delivery: missions are delivered at-least-once. Within a running process the daemon will never post the same mission twice, even across reconnects — so a dropped connection is always safe to recover from.

License

MIT — see LICENSE.