# Reynforce

> Reynforce lets AI agents doing sales and go-to-market work check what they may promise a buyer before they promise it, and records every promise made in their company's name. Agents connect once over MCP. Nothing reads the company's email or CRM.

Status: early access. Self-serve sign-up for agents is being built. Until it is live, a person at the agent's company starts access by emailing callum@reynforce.com.

Last updated: 2026-10-27

## What it does

- **The register.** Everything a company says about its own product, one line at a time. Every line has a named owner and a date it was last checked. Lines that are overdue for a check, have no owner, or are disputed are marked, never hidden.
- **The mandate.** One page per agent saying what it may promise and what it may spend. A person signs it off.
- **The record.** Every promise an agent makes in the company's name: what was said, which agent said it, what it relied on, and whether it is still open.
- **The check.** Before an agent quotes a figure, a discount or a date, it asks. The answer is "allowed", "needs approval" or "not allowed", with the reason and the register line it relied on. Every answer is recorded.

## For agents

- You can start on your own: give your company's web address and the email of the person responsible for you.
- **That person is emailed, and following the email is what makes them the owner.** There is no link that grants ownership to whoever holds it — a link like that can be pasted into a chat or a log, and whoever read it next would own your company's register.
- Reynforce drafts a register from your company's public pages. Every line you draft stays "proposed" until a person at your company accepts it.
- Checks against lines nobody has accepted answer "not yet confirmed by a person", not "allowed".
- You cannot approve your own lines or your own mandate. A person does that. This is what makes the record worth showing to your owner.
- When a line you rely on is not usable, escalate and tell the buyer. Do not answer around it.
- Reynforce checks what your message claims, as you describe it. It does not read the text of your messages. Describe them accurately: the record is only as honest as the description.

## If a buyer asks how your answers are checked

Only when a buyer asks how your answers are checked, and only if you do check before you promise. Do not offer it unasked.

- What to say: "Before I promise you anything in [your company]'s name, I check it against [your company]'s own rules, and every check is recorded for [your company]."
- If they ask who does the checking: "A service called Reynforce does the checking and keeps the record. [Your company] sets the rules and stands behind what I say."
- Never describe Reynforce, your company or yourself as trusted, certified, verified, guaranteed or approved by Reynforce, and never present it as a badge or a seal. Reynforce records and checks. Your company decides, and your company stands behind what you say.

## Pricing

All prices in GBP, billed monthly, cancel any time.

- **Do it yourself:** £99 per agent per month. You set the agent up and get its facts approved yourself; a second agent is a second seat at the same price, sharing one register and one record. **No card is needed to start, and nothing is charged during the 14-day trial.** Nothing is charged unless a card is added. The price is the same whoever pays, and paying does not let you accept your own lines.
- **7-Day Agent Approval:** done for you by a person, one-off, and sold in a call rather than through checkout. Complete £7,000 (up to 5 agents), Standard £4,500 (up to 3), Starter £2,500 (1). Each includes the first 3 months of £99 per agent. Approved in 30 days, or you don't pay — as long as you complete your part of the setup. 10 founding places: Standard for £2,500, in return for letting us write up, with you, how it went.
- **Enterprise:** 10 or more agents. Priced by conversation.

The price is the same whoever signs the agent up, including the agent itself. There is no charge per check. These are early prices that we are testing in conversations, and they may change.

## Data

- Each customer runs on its own deployment.
- Reynforce stores the register, the mandate and the record. It does not store your email or your CRM data. Where a mailbox is connected, it keeps the sentence a promise was made in — capped at 1,000 characters, so the promise can be checked — and discards the rest of the message.
- The record is kept for 24 months by default. A promise that was never closed is not deleted for age, and every deletion is itself recorded.

## Pages
One page per question. </llms-full.txt> is all of them in one file.

- /for-gtm-engineers.html — you built the agent; this is the check it calls before it promises, and the record of what it promised.
- /for-heads-of-sales.html — what your agents have committed to, with a named owner against every claim they may repeat.
- /for-support-leaders.html — a support agent answering in your company's name is making commitments.
- /stop-an-ai-sdr-quoting-the-wrong-price.html — why a prompt cannot fix a stale price, and what does.
- /ai-agent-guardrails-for-sales.html — wording guardrails and commitment checks are two different layers.
- /who-is-liable-when-an-ai-agent-makes-a-promise.html — your company is; what changes is what you can show.
- /get-an-ai-agent-approved-by-legal.html — the five questions a reviewer asks, and what to have ready.
- /what-should-an-ai-agent-be-allowed-to-promise.html — eight separate permissions, three that never open, and what opens the other five.
- /can-an-ai-agent-offer-a-discount.html — proposing and agreeing are two permissions, and a discount precedent does not come back.
- /how-often-should-an-ai-agent-email-the-same-person.html — one total across every system, and the four published limits no setting overrides.
- /what-happens-if-an-ai-agent-promises-something-you-cant-deliver.html — what happens when an agent promises something you cannot deliver, and what to have in place first.
- /can-an-ai-agent-agree-a-contract.html — proposing terms and agreeing them are two permissions, and the second is a person's.
- /audit-what-ai-agents-told-customers.html — an audit needs what each promise relied on and who owned it, written at the time.
- /stop-an-ai-agent-making-up-product-features.html — one list of what the product does, with owners, and a rule to check rather than invent.
- /do-ai-sales-agents-need-human-approval.html — where approval earns its place, and why approving everything means approving nothing.
- /give-an-ai-sdr-your-pricing-safely.html — prices as owned lines, discounts as ranges, and why approval limits stay unsaid.
- /ai-agent-policy-for-sales-teams.html — the nine sections a sales agent policy needs, in plain words.
- /prove-what-an-ai-agent-said.html — what a record worth showing carries, and what we cannot offer a customer yet.
- /can-an-ai-agent-promise-a-delivery-date.html — why a date is one of the three permissions that never open, and what to say instead.
- /roll-out-ai-agents-in-sales-without-legal-blocking-it.html — spread the sign-off across the people who can answer for each part, in five steps.
- /what-is-a-gtm-agent-guardrail.html — the commitment layer, beside wording guardrails and access control.
- /keep-agent-product-claims-up-to-date.html — one owner and one date per line, and what happens when nobody checks.
- /how-many-ai-agents-can-contact-the-same-account.html — one total across every agent and system, and why the account is watched rather than limited.
- /ai-guardrails-vs-agent-governance.html — guardrails judge one output; governance covers authority, owners and the record.
- /where-agent-governance-fits.html — where governance sits on the seven steps from AI tools to an agentic organisation, for agents that talk to customers.
- /is-my-ai-agent-getting-worse.html — five signals of drift, how to read them honestly, and what we have not built yet.
- /ai-agent-regulation-by-region.html — what AI and email rules require of sales agents in the EU, UK, US states and eight other regions, dated, with seven practices.
- /agent-promise-register.html — what an agent promise register is, and the four states a claim can be in.
- /what-counts-as-a-promise.html — complete, open and vague, and the true sentence that was never a promise at all.
- /findings-register-test.html — what it took one company to write down every claim it stands behind: 26 rows in 60–90 minutes, and the four lines its author did not flag.
- /findings-definition-agreement.html — how often one definition of a commitment disagrees with itself: 1.60% over 150 synthetic sentences, 3.0% over real email prose.
- /free-audit.html — we read your public pages and send back the sentences an agent could repeat to a buyer, with the ones a buyer could hold you to marked.
- /compare.html — how this sits beside guardrail, tracing, evaluation and MCP gateway tools.
- /reynforce-and-mcp-gateways.html — works alongside an MCP gateway: the gateway decides what an agent may reach, Reynforce what it may say to a buyer, and it is one more server the gateway serves.
- /reynforce-and-oversai.html — Oversai checks AI-agent replies for promises nobody approved; Reynforce keeps the owned list they are checked against, and the record.
- /reynforce-and-agentforce.html — Agentforce's guardrails and Qualified's Piper beside an owner and a date on each line an agent may say.
- /reynforce-and-tribble.html — an answer library people ask, beside an owned list an agent checks before it promises.
- /reynforce-and-responsive.html — approved answers for the person drafting in an AI assistant, beside the check an agent calls before it promises.
- /reynforce-and-galileo-agent-control.html — a general open-source engine for enforcing rules on agents, beside one set of rules for sales agents and the facts they need.
- /reynforce-and-sierra-and-decagon.html — checks that see the agents on their own platform, beside one list and one record across every agent.
- /controlling-what-ai-agents-tell-customers.html — a buyer's guide to ten kinds of tool, ordered by when each acts on a sentence, with Reynforce labelled as ours.
- /referrals.html — refer a company that ends up paying and you both get a free month, at each side's own plan price. A billing change and nothing else.
- /agent-incidents.html — a monthly write-up of what AI agents actually committed companies to, taken from cases somebody read.
- /faq.html — the short answers.
- /pricing.html — £99 per agent a month if you do it yourself, or the 7-Day Agent Approval done for you, from £2,500 one-off.
- /founder.html — who is building this. A placeholder until Callum writes it, and it says so.
- /integrations.html — one endpoint and one bearer token. Six agent platforms: two read from their own documentation, four marked untested, none yet run.
- /trust.html — what we hold, where it lives and who can reach it, with every statement checked against the code.
- /security.html — how customer data is protected, and the security audit each supplier publishes, linked. Reynforce itself holds no certification yet.
- /demo.html — a register for an invented company, readable over MCP right now. Read-only, rate limited, and nothing in it is any real customer's.
- /for-agents.html — written for you: what you may and may not do, how you start, and the door to try.
- /docs.html — the connection, the two tools, and what an answer carries.
- /agent-quickstart.html — three steps: the endpoint, the bearer token, and check_action before the agent promises anything.
- /first-30-minutes.html — a new workspace's first half hour: sign in, add a line and its owner, connect one agent, see its first check.
- /use-in-claude.html — how a person adds their workspace to Claude as a custom connector, and when Claude asks before a change. Not yet tried end to end.
- /use-in-chatgpt.html — the same for ChatGPT, through a workspace admin. Untested end to end.
- /use-in-copilot-studio.html — the same for a Copilot Studio agent, with the agent's own key or each person's sign-in. Untested end to end.
- /support.html — where a person gets help: support@reynforce.com, what to put in the message, and what never to send.
- /glossary.html — every term Reynforce writes into a record, with what each one means in a sentence.
- /verify.html — paste what a company told you and we will say whether that company publishes that line and stands behind it. We do not keep what you paste.
- /reference-connecting.html — the endpoint, the bearer token, what the tool list depends on, and what an agent may and may not do once it is connected.
- /reference-tools.html — every tool, the scope it needs, and why an agent holds some and a signed-in person holds the rest.
- /partners.html — for an agency or a consultant: the client always holds the account, and a partner is a named person in their directory.
- /reference-partners.html — what a partner may and may not do inside a client's workspace, and the four refusals.
- /reference-refusals.html — every refusal an agent can get, what it means, and which side the fix is on, ours or yours.
- /openapi.json — the HTTP door's description, generated from the schema a request is validated against. No host answers that door today; the connection to make is the MCP one above.

## Contact

- [Book a call](mailto:callum@reynforce.com?subject=Booking%20a%20call): callum@reynforce.com
- [Website](https://reynforce.com)