GrantKit Docs

The linter and compiler for grant proposals. Grants as files; agents bring the AI.

GrantKit is a stateless, local-first engine. It reads a grant.yaml plus your Markdown responses, then lints them, compiles them into one submission document, and reports a machine-readable status — with no cloud service and no AI calls of its own. You point Claude Code (or any agent) at the files to do the writing; GrantKit keeps them correct.

Why files¶

Agents read the full context — every section, limit, and rule is on disk.
Changes are reviewable diffs, not opaque web-form edits.
Git gives you history and rollback.
You bring the AI you already use — GrantKit itself calls no model.

It is eslint + tsc for a grant: grantkit check is the linter, grantkit build is the compiler, and funder rule packs are the config.

Quick start¶

pip install grantkit

grantkit init --funder nuffield-rda   # scaffold from a funder pack
# ...write responses/ with your editor or an AI agent...
grantkit check                         # lint against funder rules
grantkit build --format pdf --share    # compile + shareable review page

The five verbs¶

Verb	What it does
`init`	Scaffold a grant project (optionally from a funder pack).
`check`	Lint the proposal; non-zero exit on errors.
`build`	Compile responses into one document; always writes `status.json`.
`review`	Emit a review packet for an AI agent (no AI calls).
`status`	Completion %, per-section word counts, deadline countdown.

Next steps¶

Created by PolicyEngine