Prices a scope from your rate card in code, then has the AI write the SOW prose around the computed figures and the discovery notes. The model never prices anything and never sees a rate it can change.

Provided free and as is, without warranty of any kind — including merchantability, fitness for a particular purpose, and the accuracy or completeness of any result. See the licence. You are responsible for checking what this produces before relying on it.

Proposal Drafter

The most-hated recurring task at a firm under twenty people, reduced to reviewing a draft.

What it does

Reads two tabs — RateCard and Scope — prices the scope line by line in python, and prints a phase-by-phase table with subtotals and a total. Then it hands that finished table, plus your discovery notes, to the AI, which writes the surrounding SOW: background, objectives, scope of work per phase, out-of-scope, assumptions and dependencies, and an investment section that restates the computed figures.

The message carries the draft first and the pricing table underneath it.

Where the money is computed

In the compose task, in plain python, before the model is called. The AI receives a table it cannot change and is instructed not to produce any figure that is not already in front of it — no arithmetic, no estimates, no invented rates.

This is the whole design. A model that prices your work is a model that will eventually underprice it, in a document you send to a client with your name on it. Delete the narrate task and you still get correct pricing; you just have to write the prose yourself.

A role that appears in Scope with no matching row in RateCard is priced at 0 and called out by name at the bottom of the table. It is not guessed at, and the AI is told to say so plainly rather than smooth over it.

What it does NOT do

This produces a draft. It is not a contract, it is not legal advice, and it is not sendable as-is. It gets you past the blank page, which is the part that takes the week.

It does not know your standard terms, your insurance position, or what you agreed verbally on the call. Out-of-scope in particular deserves your eyes every time — it is where firms lose money, and the model can only infer it from notes you wrote.

Setup

1. Google Sheets

Two tabs in one spreadsheet.

RateCard:

role	rate

Scope:

phase	role	hours	notes

Role names are matched case-insensitively between the two. Keep them consistent or you will get a 0-priced line and a callout.

2. Discovery notes

A plain-text or markdown file the InTouch server can read. The first 8,000 characters are used — enough for a long call, short enough that the model stays on the brief.

3. Credential + contact

Replace the placeholders: - <<GOOGLE_WORKSPACE_CREDENTIAL>> — Google Workspace credential with Sheets read scope - <<PROPOSAL_SHEET_ID>> — the sheet holding both tabs - <<DISCOVERY_NOTES_FILE>> — path to the notes - <<ANTHROPIC_CREDENTIAL>> — used by the narrate task only - <<PROPOSAL_OWNER_PUBLISHER>> — who receives the draft

4. Schedule

On demand. Run it after a discovery call, not on a timer.

Customization

House structure. The section list lives in the narrate prompt. Replace it with your own SOW skeleton — that is the single highest-value edit here.
Local model. Swap anthropic for ollama and the notes never leave the building. Only the narrative step touches a model, so it is a one-task change.
Drop the AI. Delete narrate and set the message body to {{compose.output}} for pricing only.
Discounts and fixed fees. Add a column to Scope and apply it in compose, so the adjustment is still computed rather than described.

Source

workflow.iml

AI provider

Any supported AI provider works here. This template ships configured for the anthropic tool; swap that task's tool to anthropic, openai, gemini, mistral, ollama or openrouter to use the provider you already have, and point its credential at your own key. The step is a plain summarise/classify call — nothing in it is vendor specific. See requires.ai in the manifest.