DavidAgents

DavidAgents — Praetor

The spine · outbound compliance

Every call. Every text. Every decision, proven.

A deterministic compliance engine in front of every outbound contact the platform makes — 51-jurisdiction calling rules, a consent ledger, instant opt-out enforcement, and a tamper-evident audit trail. No LLM in the decision path: same input, same answer, every time.

In Rome the praetor published the edict that set what the republic would permit, then ruled by it. It belongs beside Michelangelo's David — the ideal made real, and the magistrate who guards its conduct.

Put a gate on my outreach See it running

The problem

Agents that reach out are agents that carry legal risk.

TCPA suits start at $500 per violation and carriers suspend first, ask later. Most platforms bolt compliance onto the campaign tool and hope nothing else sends. But the risk lives in everything else — the agent that texts a follow-up at 8:10am into California, the callback that fires on a Sunday in Alabama, the number that said STOP last month. So we built the gate we could not buy, put it at the carrier door where nothing can route around it, and turned it on for every line we sell.

Here's what every page on this site is really about: this whole company — every product line on it — was built and is run by one founder and his agents, on the platform you're reading about. That team is the proof it works. What we sell is the same leverage, priced so anyone can have it — because how good you are should decide how far you go, not your software budget. Why we build this way →

Capabilities

The gate

One deterministic door for every outbound contact

Before any SMS or dial leaves the platform — agent-initiated or human-initiated — the sender asks the engine. It runs every check and returns allow, review, or block with the statute behind each ruling. Consumer STOPs are honored instantly on every line and suppress the recipient platform-wide: a hard block that nothing overrides, not even an allow-list.

● a suppressed number is refused before the carrier is ever called — re-proven on our live test wall

51 jurisdictions

State law is the rule, federal is just the floor

Twenty-three states enforce stricter calling windows than the federal 8–9; eight close Sundays or holidays; forty-three carry their own mini-TCPA statutes, most with a private right of action. The engine resolves the recipient's local clock and applies the tightest law — each rule citation-backed and adversarially machine-validated, with the contested ones held for counsel rather than silently allowed.

● all 51 jurisdictions re-swept 2026-08-20 — every change verified against the primary statute

Consent & caps

The paper trail and the throttle, built in

A consent ledger with proof, source, expiry, and revocation; established-business-relationship windows applied exactly where the law honors them; per-number and per-campaign velocity caps over rolling windows; SHAFT content screening before a carrier ever sees a message. Your own business rules ride the same engine — as data, evaluated safely, never as code.

● quiet-hour sends reschedule themselves to the next legal minute

Evidence

Compliance you can hand to a regulator

A policy document says what you intend. A hash chain proves what you did. Every decision links to the one before it — alter any record and the chain visibly breaks. When someone asks for every contact decision on a number, that is one export, regulator-ready, not a discovery project.

● the audit chain verifies live from our portal, every day

iMessage

The only compliance engine for the blue bubble

iMessage has no carrier regulation — no 10DLC, no campaign vetting — and every vendor stops the sentence there. We finish it: Apple bans senders without appeal, and any "unregulated" send to a phone number can silently fall back to carrier SMS, where the $500-per-message statutes attach in full. So the engine evaluates every blue-bubble send at its worst-case channel — the complete 51-state text ruleset — layered with the platform-survival governance nobody else sells: per-line new-conversation velocity, fresh-number warm-up ramps, no-reply chains, cold-first-touch gating, blast fingerprints. A STOP on any text channel suppresses them all.

● a suppressed number is refused on iMessage before the provider is ever called — proven on the live wall

The living edict · swept 2026-08-20

What actually changed in three weeks.

In Rome the praetor republished the edict as the law moved — the rules he would enforce, posted in public. This is ours. Three weeks after the last republication, we swept the Federal Register, the circuit courts, the FTC, the numbering authority, and all fifty-one jurisdictions again — deltas only, primary sources only. Then an independent pass attacked every finding, instructed to refute it from the statute itself. Forty-four survived. One died — and the one that died is the best part of the story.

51jurisdictions swept
58research + verify agents
44findings confirmed
1refuted — and reversed
1day, FCC notice → tracked
  1. Aug 20 corrected

    Alaska's consent exception doesn't exist. Our own entry said it did. We publish that.

    Our committed rule let a prerecorded pitch into Alaska if a live operator got permission first — an exception the legislature repealed in 2004 that survives only in stale summaries, including, until today, ours. Two independent agents re-derived the statute from Alaska's own legislative database: the ban on automated and recorded sales messages is unconditional, no consent cures it, and it carries a private right of action of treble damages or $500 per call. The gate now blocks where it used to pass. A compliance vendor that never corrects itself in public isn't verifying anything.

  2. Aug 20 verified

    Our scanner tried to close a watch item. Its own verifier proved it wrong.

    The research pass concluded a flagged FCC lobbying letter was routine paperwork and moved to clear our watch on the revoke-all consent rule. The adversarial verifier — instructed to refute, not to agree — pulled the actual letter from the filer's own site: it is a live industry campaign to narrow how consumers revoke consent, with a request to delay the rule a further twelve months. The watch escalated instead of closing. That is why every finding faces a hostile second pass before it touches the gate — including the reassuring ones. Especially the reassuring ones.

  3. Aug 20 gap closed

    Washington D.C. flatly bans synthesized-voice sales calls. Now the gate does too.

    D.C. Code § 34-1701 prohibits automated calls delivering a prerecorded or synthesized voice for commercial solicitation — language written before AI voice existed that reaches it perfectly. Our baseline carried no D.C. entry beyond the federal floor. It does now, as a hard stop no consent overrides, alongside the district's telephone-fraud statute and its private right of action.

  4. Aug 20 gap closed

    West Virginia's telemarketing act was a promised follow-up. Promise kept, encoded, dated.

    The July sweep found West Virginia's $100-to-$3,000-per-violation private right of action but held it for a dedicated verification pass rather than rushing it in. This sweep ran that pass: the full article — abusive-acts list, safe harbor, cure window, bona fide error defense — is now in the ruleset with corrected pin-cites for the calling window our old entry had mis-cited. We'd rather ship a gap with a date on it than a guess with a checkmark.

  5. Aug 20 data

    At 6pm UTC, San Juan is mid-afternoon and Guam is 4am. The gate now knows both.

    The five U.S. territories were missing from our area-code map, so a call to Guam resolved to no timezone at all and fell through to a warning. Federal law reaches the territories exactly as it reaches states — and the timezone spread is the whole risk: one dispatch that is legal in Puerto Rico is a 4am violation in Guam at the same instant. Mapped, blocked, and pinned by a test. Two missing Canadian area codes came back in the same pass, so Canada's opt-in regime applies again where it had silently lapsed.

  6. Aug 19 watch

    The FCC moved to dismiss 24 old TCPA petitions. It was in our watch ledger the next morning.

    A Bureau notice released August 19 proposes dismissing, with prejudice, twenty-four petitions filed between 2003 and 2023 — including one touching the reassigned-numbers database our safe-harbor check rides on. No rule we enforce depends on any of them, so nothing changed in the gate; but the dismissal clock, the objection window, and the one petition conspicuously absent from the list (the pending quiet-hours petition) are all recorded, dated, with a recheck set before the window closes.

The next docket is already committed inside the ruleset itself: the FY2027 do-not-call fee schedule due by September 1 with an increase statutorily locked in, a North Carolina consent rewrite alive in conference through December, a Georgia area code about to get its in-service date, and the revoke-all rewrite the refuted finding forced us to keep watching. When the law moves, the edict republishes — and you'll read about it here, dated.

Manual outbound dialing · the half nobody gates

Someone on your team just typed a number and pressed call.

Every compliance vendor gates the campaign — the blast, the sequence, the scheduled drip. None of them gate the salesperson who pulls up a keypad and dials. That call has no list behind it, no approval step, no send window; it goes from a browser to the carrier in one hop, and it carries exactly the same $500-to-$1,500-per-violation statutes as the campaign that took three days to get approved. It is the most common outbound call a company makes and the least governed. So we put the same door in front of it.

  1. A gate that refuses everything is a gate people route around.

    This is the part that makes manual dialing hard, and it is where a naive filter does real damage. A human dialing and speaking is not an autodialer — the federal written-consent rule is built on automated and artificial-voice calls, so demanding consent before a live dial would refuse the vendor callback, the returned voicemail, the B2B prospect who asked you to ring back. Consent is advisory here, and the engine says so on the record. What still binds absolutely: a recipient who said STOP, the calling window in their state, the mini-TCPA rules, the do-not-call posture. Enforce the rules that apply; get out of the way of the ones that don't.

  2. It refuses in the recipient's local time, not the caller's.

    The screen here is 8:54pm on the east coast, and Florida is an hour stricter than the federal floor — its solicitation window closes at 8:00, not 9:00. A federal-only check would have connected this call. The engine resolves the recipient's local clock, applies the tightest law that reaches them, and treats a number it can't place with less confidence rather than more. The same dial to a California number at that instant connects, because in California it is 5:54pm and legal.

  3. The person who gets refused is the person who has to explain it.

    So the refusal carries the statute, on screen, at the moment of the dial — not a generic error and not a line in a log someone reads next quarter. The operator can see which rule fired, read the citation behind it, and answer the question their manager is about to ask. Every refusal also carries its audit identifier into the hash chain, so the same event is retrievable months later without anyone reconstructing what happened.

  4. An outage is not permission. It fails closed.

    The tempting design is to let calls through when the engine can't be reached — nobody's phones go down, and you log the gap. We rejected it. An outage is precisely the moment the record goes blank, and "we called four hundred people during the window our compliance engine was down" is not a sentence anyone wants read back to them in a deposition. If the gate cannot answer, the regulated call does not go, and the operator is told why in the same words as any other refusal.

  5. But the passcode still sends. Failing closed is scoped to where the law applies.

    A gate that kills two-in-the-morning signing codes gets switched off within a week, and then nothing is gated at all — so the carve-out has to be principled rather than convenient. Ours is the engine's own taxonomy: the statutes here govern solicitation, and every regulated check already passes a passcode the recipient asked for seconds ago or a reply to their own message. For those a healthy gate's answer is knowable, so an outage changes nothing. For a cold marketing call it genuinely isn't, and unknown has to mean no. One rule that a suppressed number is refused holds through all of it — no outage carve-out has ever reached a recipient who said STOP, and a test proves it on every commit.

We did not build this from a threat model. We built it because our own dialer talked straight to the carrier and nothing stood in front of it — an operator could have called a suppressed number at 10pm on a Sunday and the platform would have helped. That is the gap in most stacks right now, and it is invisible until it is a filing.

What's inside

The manifest.

  1. 01 Deterministic gate allow / review / block — no LLM in the path
  2. 02 51-state ruleset citation-backed, machine-validated
  3. 03 Opt-out enforcement STOP honored instantly, platform-wide
  4. 04 Consent ledger proof, basis, expiry, revocation
  5. 05 EBR windows purchase 18mo / inquiry 3mo, where honored
  6. 06 Velocity caps per-number, per-campaign, rolling windows
  7. 07 Content screening SHAFT + carrier-filter risk, pre-send
  8. 08 iMessage governance worst-case-channel gate + Apple survival rules
  9. 09 10DLC / STIR-SHAKEN campaign + attestation posture per number
  10. 10 CAN-SPAM · CASL · GDPR email + international headline regimes
  11. 11 Custom business rules your policy as data, safely evaluated
  12. 12 Hash-chained audit tamper-evident, regulator-ready export
  13. 13 Quiet-hour rescheduling blocked sends move to the legal minute

How it lands

Already on

Every AgentsFast plan ships with the gate live: opt-outs enforced, federal quiet hours applied, content screened. There is nothing to install — outbound simply cannot skip it.

Go enforcement-grade

Compliance Plus flips your business to enforce mode: the 51-state rules block instead of warn, the consent ledger and velocity caps come alive, and the evidence trail is yours to export. One flat $99/mo — your agent turns it on in chat.

Bring your counsel

On DavidAgents, Praetor runs inside your own instance with your counsel's rulesets layered onto ours — compliance isolation, data residency, and the same provable gate wherever your agents run.

Proof · the compliance framework

We show you where the law is unsettled.

Most compliance tools sell you a green checkmark and a confidence you can't audit. Ours is built the opposite way — every rule traced to primary law, adversarially attacked before we trust it, and honest about the exact places the ground is still moving.

  1. 01

    Research the source

    Every rule is traced to primary law — the eCFR text of 47 CFR 64.1200 and 16 CFR 310, the state statute itself, the FCC orders and the court opinions — never a vendor's summary of a summary.

  2. 02

    Attack it

    A second pass tries to REFUTE each rule: does the statute actually reach texts, or only voice? Is the citation the right subsection? Did a 2024–2026 amendment or ruling quietly change it?

  3. 03

    Grade the certainty

    What survives gets a tier — SETTLED (clear statute + cases), LIKELY (reaches it, thin authority), or CONTESTED (a genuine live split). The ruleset is flagged validated only when counsel-confirmed.

  4. 04

    Gate by certainty

    The engine acts on the grade: block on settled, review on likely, advise on contested. No LLM in the decision — same input, same ruling, every time, with the statute printed behind it.

The validation ledger — what's settled, and what isn't

51 US jurisdictions · state calling rules 33 counsel-validatedsettled
The 7 formerly-contested states (CA·IL·ME·MA·MS·NV·SD) resolved 2026-07settled
Re-opened by the 2026-07/08 sweeps (PA·WA·WY·TN·MS·AK·DC·WV·RI·MD) encoded · queued for counsellikely
Federal TCPA / TSR — 6 of 7 rule blocks validatedsettled
Federal — autodialed wireless written consent heldcontested

⚖ Why the federal ruleset still says "pending sign-off"

In February 2026 the Fifth Circuit (Bradford v. Sovereign Pest Control), riding the Supreme Court's McLaughlin decision, held the FCC's prior-express-WRITTEN-consent rule for wireless marketing unenforceable in Texas, Louisiana and Mississippi. It's a live circuit split. So we keep the strict written-consent floor in the gate — but we refuse to stamp the federal ruleset "validated" while the law is genuinely contested. We'd rather show you the edge than hide it.

183 deterministic engine tests — compliance checks live on the wall — failing last run —

Certainty where the law is settled. Honesty where it isn't. For a deterministic gate that decides whether your message is legal to send, that distinction is the whole product.

See the live gate All the receipts

Put Praetor on your payroll.

Same platform we run our own company on. First line is a purchase — the second is a decision.

Put a gate on my outreach