Trust · The Pledge

The Grounded-AI Pledge.

Marketing pledges rot. Contracts don't. This pledge is the contractual section of every PursuitAgent MSA — not a blog post.

Version v1.0 · Effective 2026-04-20

contract sections
6

version v1.0, effective 2026-04-20

change notice
90d

for material protection reductions

unsupported claims
0

drafting must cite approved sources

Contract control

What the pledge forbids

No training use

Customer documents never train PursuitAgent or third-party models.

No unsupported drafting

Every generated answer must trace to approved KB content.

No silent source expansion

Web search, other tenants, and model memory stay out of drafting.

  1. 1 · Grounded responses only

    PursuitAgent never generates proposal responses from content outside your approved knowledge base.

    Every AI-drafted answer in PursuitAgent is retrieved from — and cites back to — a specific content block in your KB. The underlying models are never asked to "write" from their training data; they are asked to compose citations of content you own.

  2. 2 · Mandatory citations

    Every AI-drafted sentence links to the exact KB source + page.

    Citations are rendered inline at the sentence level, not at the paragraph or section level. Clicking a citation opens the source document at the cited page. If the KB cannot support an answer, PursuitAgent returns "not enough info" rather than fabricating one.

  3. 3 · Restricted sources

    Drafting uses your uploaded proposals, your approved KB content, and the specific RFP you're responding to — nothing else.

    Sources PursuitAgent will never silently use for drafting: the model's training data; other customers' KBs; web search results; third-party knowledge graphs. Any change to this restriction is governed by §6 below.

  4. 4 · No training on your data

    Your uploaded documents, drafts, and KB content are never used to train any model — ours or a third party's.

    PursuitAgent contracts with model providers (Anthropic, OpenAI, Google) that explicitly forbid training on API inputs. We re-validate this quarterly; the audit is available to customers under NDA.

  5. 5 · Termination right

    If PursuitAgent violates §1–§4, Customer may terminate without penalty and recover pre-paid unused fees.

    Violation means a drafted answer that cannot be traced to an approved KB source, a draft whose grounding claim is materially false, or documented use of customer data to train any model. Customer must give PursuitAgent 15 business days to cure; if not cured, termination is effective on notice.

  6. 6 · Change control

    Material changes to this pledge require 90 days' advance notice and Customer may terminate if the change reduces protection.

    "Material" means any change that expands the allowed sources in §3, weakens the citation requirement in §2, or narrows the termination right in §5. Non-material clarifications (typos, examples, footnote updates) can ship without notice; they are tracked in the changelog below.

Changelog

  • 2026-04-20 v1.0 — Initial pledge. Six sections. Published as contractual in all MSAs signed on or after this date.

See the pledge in action.

Start a trial workspace and every generated answer must cite the approved source material it came from.