Open source · MIT

Discovery calls in.
Branded proposals out.

A pipeline that turns a meeting transcript into a signature-ready Quote + SOW in under five minutes. Schema-validated end to end, eval-driven, self-hosted on your stack. Built for AI agencies.

Star on GitHub →

Transcript

[12:34]So what we're looking at is automating the entire intake process. Right now we lose about twelve hours a week just on follow-ups…

[14:18]I'd estimate a $5K pilot followed by a monthly retainer. Send the proposal Friday?

Proposal · v1

Acme Co · Quote + SOW

Pilot$5,000

Monthly retainer$2,000/mo

Discovery + scopeIncluded

12 pages · ready to send

The loop

From transcript to signed proposal.

Drop the transcript

Paste it in or wire a webhook from Read.ai, Fathom, or Fireflies. Signature-verified, idempotent.

The pipeline does the work

Extract → pricing-readiness gate → generate → critique → render. Schema-validated end to end, retried on stochastic failures.

Edit, send, track

In-browser preview, mailto handoff, six e-sign providers for the close. Failure modes surface inline.

Integrations

Connects to what you already use.

Three meeting recorders for inbound transcripts. Six e-signature providers for the close. Add yours via the same provider abstraction.

Meeting recorders

Read.aiSupported
FathomSupported
FirefliesSupported

E-signature providers

DocuSignSupported
PandaDocSupported
SignWellSupported
BoldSignSupported
OpenSignSupported
Wet SignSupported

Under the hood

Built for engineers who hate proposal slop.

Schema-validated LLM output

Every section parses against a Zod schema before render. Failed parses retry with a temperature nudge; persistent failures fail the run loud, not silent.

Eval-driven CI gate

Golden-set evals run on every PR. Regression beyond 5% fails the build. Prompts ship only when they win pairwise.

Adversarial review workflow

Multi-model challenge on every diff. Cloud reviews surface P1 issues before merge — no rubber stamps.

Operator-tunable templates

Your tiers, your prose, your terms, your discovery questions. Settings replace generic defaults; per-vertical fallbacks fill the gaps.

427

tests

integrations

API routes

<5min

transcript → PDF

Next.js 16React 19InngestSupabaseVercel AI SDKAnthropicZodLangfuseReact-PDF

Get started

Clone, configure, ship.

Terminal

# Clone
git clone https://github.com/janderswag/proposal-agent.git
cd proposal-agent

# Install + run the eval suite
npm install
npm run eval

# Deploy to Vercel
vercel deploy --prod

View on GitHub →

Discovery calls in.Branded proposals out.