Target Radara StoryC Ltd product · no tenant yet
A StoryC Ltd product · installed in your own Supabase and GitHub projects

A short list with a reason and a link. Never an approach.

A commercial team rebuilds the same target list from memory and last week's press. Target Radar files a short list of proposed companies from the sources you already pay for, each with the claim it met and the link a named person can open. It does not write the approach and it does not email anyone. The introduction stays a person's.

The model proposes

Your own model, on your own key, called only after a fixed rule has scored the rows. It may not add a name, reorder, invent a link, name a fee or draft an approach; the code withholds it when it tries.

The rules dispose

A claim scores only with a pointer from a connector: a link or a register record the person can open, and the words that matched. No pointer, no points. No points, not filed.

A person signs

Three signatures before the routine arms: the search profile approved, at least one source linked on your licence, the cost estimate confirmed with a cap. Then a named person approves or declines each name.

Your tenant

Open an existing tenant

The key is sent only to your own function, at the address in config.json.

Create the tenant

The first tenant on a fresh install needs no install key; after that the install key set in your dashboard.

1 · The interview

One question at a time, in your words. A vague answer gets a narrower question; nothing is invented from it. No prompt box exists here: the questions are a fixed list.

…

2 · Your page, then a proposed profile

Paste the address of the page a new customer would read. The function fetches that host only, public https only; nothing else is searched and no paid connector is called. The page is read as data: an instruction found on it is shown to you and not followed.

3 · Link your own connectors and your own model

Each connector is a slot with a contract. The vendor is your choice, on your own licence, with your own key; nothing is linked on your behalf and nothing is linked before the profile is approved. Before each link you see what it will be allowed to do and what it will never do. A grant that asks for a send scope is refused.

4 · The cost, before the routine is armed

No vendor price is invented. For each linked connector, type what one lookup costs on your plan, or mark it as included. The estimate uses only the figures you confirm; the first paid call happens after you confirm it, not before. It is not a quote from StoryC; the vendor's invoice is yours.

The estimate

5 · Arm the routine

Only when the three signatures exist. Daily or weekly, your timezone, your hour, from the interview. Output goes to the named person only. On demand runs from this console are allowed after arm.

The three signatures

The routine

inside the caps; the digest lands in the queue

The named person's queue

A draft, not a send. At most five proposed targets in score order, each saying who they are, why they fit using only the signed claims, and the locator for each claim; one line on what was left off. Status stays proposed until the named person decides. A name added by hand is manual and unscored.

Every proposed name

CompanySourceScorePointersStatus

Receipts

Every run leaves a receipt: the agent, the prompt key and version, the model name or none, the input and output hashes, the gates, the lookups against the cap, and the reason anything was withheld. Pointers and span hashes, never a page body.

WhenRunPromptModelGatesCap usedWithheld

The kit: one build, many clients

Target Radar is installed into your own Supabase project and your own GitHub repository. A second client on the same build sees nothing of the first: every row carries the tenant, row-level security is on every table with no open policy, and the function scopes every read and write to the tenant it resolved from the key.

What is in it

  • the migration: tenant, answers, profile, grants, costs, routine, targets, declines, queue, receipts
  • the function: the interview, the own-page fetch, the card, the links, the cost screen, the arm gate, the run, the decisions
  • the fixed rules as one file, run in the browser, in Node and in the function
  • the locked prompt target-radar.v1, never edited; a change is a new version
  • the acceptance tests and the self test the database runs on itself

Open the kit on GitHub

What it never does

  • no prompt box, no editable prompt; the questions are a fixed list
  • no scraper; your own page only, public https only, no private or metadata address
  • no vendor baked in; every connector is empty until you link it
  • no send, reply, forward or delete on any mailbox grant
  • no price, fee, discount or commitment in a digest
  • no name the rule did not qualify; no instruction from a page followed
  • no routine armed before the three signatures; no invented unit price

Self test on this install: