Skip to content

Proposal Generation

Start a proposal from a pursuit: open it from the Pursuits list, go to its Proposal tab and choose Start a proposal. The run starts from that notice’s own text. It reads the solicitation, plans the required sections, writes each one and reviews its own work. (The separate Generate Proposal page was retired on 2026-10-10.)

This produces a draft, not a submission-ready document. Every section is meant to be reviewed by a person before it goes out. The platform flags anything it is not confident about rather than guessing silently.

AI drafting is off in production. The platform makes no live AI calls for customers at present. A proposal run still completes its steps, but each section holds fixed sample text and is labeled as such: “AI drafting is off in this environment. This section holds fixed sample text, not a draft for this notice.” You write the section yourself with Edit, or run the proposal again once AI drafting is on. A sample section cannot be approved. Nothing that looks like generated writing is passed off as real.

A LangGraph state machine runs parse -> compliance -> plan -> write -> review -> assemble, with error edges that stop the run and record the failure. The run and each section are saved to the database as the graph progresses.

When AI drafting is on:

  • Each section lists the excerpts from your own Knowledge Base documents that the writer was given. They are not citations to the solicitation.
  • The review step checks the section’s claims against those excerpts. Anything it cannot support is listed as an unsupported claim, and a section below the confidence bar is flagged for human review.
  • Regulatory clause citations are grounded against the FAR/DFARS clause catalog, see Clause Library.
  • You approve or edit each section. There is no reject control.

Try this in the platform