Find the conversations worth answering
Run a scan whenever you want. Exa searches the open web and Hacker News answers alongside it, and every discussion that comes back is scored for whether you would be a legitimate voice in it. Replies are drafted in your tone, checked, and staged for you to send by hand.
“Has anyone got this working without the paid tier?”
What Entry Point does with a thread
Three scores, weighted 40/40/20
Entry legitimacy asks whether you are the right voice in this thread. Value asymmetry asks whether a reply would actually help the person who posted. Social risk runs the other way, where a higher number means a reply is safer to make. The three combine at 40, 40 and 20 percent against a bar you set, 50 by default. A thread that misses that bar by up to twelve points still reaches you when it carries real engagement, and nothing scores in below 35.
Four flags, and one of them can abstain
Each thread is marked for an explicit request for help, detectable confusion, misinformation worth correcting, and a question still unanswered. That last flag needs a source that reports its reply count. A web page found by Exa reports none, so the flag stays off and the thread does not collect points it has not earned.
Three drafts, in your voice, in different approaches
Each opportunity gets three drafts by default, each taking a different approach: reframing the problem, correcting gently, synthesizing what the thread already said, answering directly, or sharing what happened to you. They are written against the brand voice and tone you set in settings, the learnings you have saved, and any style rules you have written for this site.
A second pass that can refuse the draft
The model is told not to include a link, a product mention, or a call to action. Then a checker that does not involve a model reads what it wrote and rejects the draft outright if it finds a URL, promotional phrasing, harmful content, or language implying an authority you do not have. Professional advice and overlong replies come back as warnings. The verdict is stored with the draft, so you can see what was caught.
What a scan searches today
Entry Point ships with four sources and reads each one’s availability from the running environment. Two of them answer here, and the settings panel tells you the same thing it tells you below.
Neural search across articles, forums, changelogs and documentation, pulled back with the page text so a thread can be scored on what it actually says.
Full-text search over posts and comments, with points and reply counts, which is what lets the unanswered-question flag do its job.
Reddit and X are built into the product and are not searchable on this deployment.
Reddit refuses anonymous searches from a server address, and both sources need connection credentials that quietly.sh does not currently hold. Nothing in the app pretends otherwise: the settings panel marks each one unavailable and names the missing credential. If that changes, the toggles start working and the scans include them.
How it works
- 1
Describe who you help
Write down what you do, who it is for, how you sound, and what you can speak to with authority. Add the topics you want nothing to do with and threads touching them are rejected while they are being scored, before anything is written.
- 2
Run a scan whenever you want
Start a run on demand and watch it work: the search fans out, results are stored, and each conversation is scored against your context and your saved learnings. What clears your bar lands in the inbox, and what does not is counted against the reason it was rejected.
- 3
Reply, stage, or archive
Open a thread, read the drafts and the reasoning behind each, and edit before you send. Push the inbox into Fabric as rows with reply and outreach columns when you want to work through them as a batch, mark a run reviewed, and archive what is not a fit. Learnings you save are read back on the next scan.
Find the threads worth answering
Seven-day trial. Run one scan and read the scores before you write anything.