OpenGrantIndex

live Demo: https://ogi.shakespeare.wtf

An open-source, decentralized search engine for grants, fellowships, bounties, RFPs, prizes, hackathons, accelerator funding and public-good funding.

Google for grants — but open, searchable, API-first, community maintained, and structured so that nobody, including us, can lock it up later.

A project of the NAI Institute. MIT licensed.

---

The problem

The money is public information. Foundations publish their calls, governments run portals, universities post fellowships, open-source ecosystems announce bounties, and IRS Form 990 filings disclose every award a US private foundation has ever made.

And yet finding the right funding is still weeks of work. The information is spread across thousands of sites with no shared schema and no shared notion of a deadline. Existing aggregators each cover one slice — historical filings, or AI safety, or EU research, or crypto public goods — and the good ones eventually become products with paywalls, or quietly disappear.

Nothing continuously indexes active funding opportunities across all of these worlds at once, and nothing joins those live calls to the historical record of who actually got paid.

What makes this different

Decision	Why
The corpus is signed Nostr events, not rows in a database	Anyone can mirror the whole index for the cost of a VPS. Shutting it down is not an action we are capable of taking.
The canonical URL is the primary key	Two crawlers written by strangers derive the same identity for the same call, so deduplication needs zero coordination.
Trust is computed in your browser	We publish signals and a formula you can read. Change your trust roots and the same corpus ranks differently.
Disagreement is displayed, not resolved	When publishers report different deadlines you see both, who said what, and why one ranked higher.
Open calls are joined to historical awards	A funder's priorities page is marketing; its award history is revealed preference.
The frontend works with no backend at all	Search, dedup, trust scoring and graph queries all run client-side against relays.

Architecture

Sources (foundations, gov portals, GitHub, RSS, Nostr, community)
   │
   ▼  discover() → crawl() → extract() → normalize() → validate() → publish()
   │
Nostr relays  ◀── source of truth (kinds 35231 / 31457 / 34011 / 37063 / 9987)
   │
   ├── Static frontend (relay-direct, zero backend)   ← this repo
   ├── REST + GraphQL API (optional)
   └── Postgres mirror + Meilisearch (optional)

Every component except the relays is disposable. The failure mode of a conventional grant database is total data loss; here it is staleness.

Full details: /architecture covers the plugin contract, crawler framework, AI extraction ladder, deduplication rules, database schema, search schema and repository layout.

Event kinds

Defined in NIP.md.

Kind	Record	Class
35231	Funding opportunity	Addressable
31457	Funder profile	Addressable
34011	Historical award	Addressable
37063	Crawler source manifest	Addressable
30441	Saved search / alert	Addressable
9987	Opportunity attestation	Regular

Everything else reuses an existing NIP rather than reinventing it: discussion is NIP-22, labels are NIP-32, bookmarks are NIP-51, external identity is NIP-73, listing tag conventions borrow from NIP-99, and deadlines mirror to NIP-52 calendar events.

Getting started

npm install
npm run dev      # http://localhost:8080
npm test         # typecheck + lint + tests + build
npm run build

The app ships with a bundled corpus snapshot so it is fully functional on first paint, before any relay has answered. Live signed records from relays always take precedence over the snapshot.

Automation

The site keeps itself fresh without any manual curation:

• scripts/crawl/ — a TypeScript crawler (npm run crawl) that polls every enabled source declared in src/lib/ogi/seed/sources.ts. RSS/Atom feeds and the Grants.gov Search2 API are supported today; sources whose endpoints need html/sitemap/graphql/pdf adapters are skipped and recorded as such. Results are normalized with the exact identity rules of the frontend (canonicalizeUrl is the primary key), merged with the previous snapshot — records that vanish from a source for over 30 days, or whose deadline passes, are marked closed — and written to src/lib/ogi/seed/generated.json, generated.ts and crawl-report.json.
• .github/workflows/crawl.yml — a cron workflow (every 6 hours, plus manual dispatch) that runs the crawler and commits the refreshed snapshot. The push triggers the deploy workflow, so the site auto-updates.
• Relay publishing — when the repository secret OGI_BOT_NSEC contains an nsec key for the crawler bot, the run also publishes signed kind 35231 events (same tag layout as the snapshot) to the app relays. Without the secret the crawler runs in snapshot-only mode and the workflow still succeeds.

Useful flags: npm run crawl -- --source=<id> crawls one source, --dry-run prints without writing, --publish forces relay publishing.

SIP-01

The crawler doubles as a SIP-01 indexer node. In publish mode, each crawled opportunity produces three things:

1. kind 35231 — the OGI opportunity record (the corpus itself),
2. kind 39697 — a SIP-01 web-index observation of the opportunity's page, with identity derived by the vendored protocol core (scripts/crawl/lib/sip01-webIndex.ts, copied verbatim from sip-01-core/src/protocol/webIndex.ts; SIP-01 §13 byte-compatibility is enforced by scripts/crawl/lib/sip01-webIndex.test.ts). SIP-01's normalizeIndexUrl intentionally differs from OGI's canonicalizeUrl — the two identities are computed independently and never mixed,
3. kind 16919 — one replaceable heartbeat per run (source: ogi-crawl/1), reporting pagesIndexed/published stats so the SIP-01 network can see the node is alive.

Per SIP-01 §14, OGI_BOT_NSEC must be a dedicated bot keypair — never a personal key; the pubkey is the indexer identity and rotating the key simply creates a new indexer.

Project layout

src/
├── lib/ogi/
│   ├── types.ts        canonical domain model
│   ├── kinds.ts        event kind constants
│   ├── normalize.ts    URL canonicalization, deadline & amount detection
│   ├── parse.ts        Nostr event → domain object
│   ├── trust.ts        trust scoring + deduplication/merge engine
│   ├── search.ts       BM25 index, NL query parser, facets, URL state
│   ├── graph.ts        knowledge-graph query engine
│   ├── export.ts       RSS, iCal and JSONL generation
│   ├── format.ts       amount/date/country display helpers
│   ├── routes.ts       canonical-URL permalinks
│   └── seed/           bundled corpus snapshot
├── hooks/
│   ├── useOgiIndex.ts          one relay query → merged, trust-ranked corpus
│   ├── useOgiSearch.ts         URL-driven search state
│   ├── useOgiPublish.ts        submit opportunities, publish attestations
│   ├── useSavedOpportunities.ts NIP-51 bookmark set
│   └── useSavedSearches.ts      kind 30441 saved searches
├── components/ogi/     Layout, cards, filters, attestations, provenance
└── pages/              Home, Search, Opportunity, Funders, Graph, Awards,
                        Sources, Submit, Saved, and the doc pages

Contributing

Four useful things, in increasing order of effort:

1. Submit an opportunity (2 min) — publishes a signed record attributed to you, live immediately.
2. Attest to a record (10 sec) — confirm a grant is open, report a dead link, correct a deadline. This is the highest-leverage contribution, because it is what keeps a decentralized index fresh.
3. Register a source (30 min) — publish a kind 37063 manifest and any worker can crawl it.
4. Write a crawler plugin (an afternoon) — five methods against a typed contract.

Licence

Code: MIT. Normalized metadata: CC0. Original source text remains under whatever licence its publisher applied, recorded per-source in the kind 37063 manifests.

---

_{Vibed with Shakespeare}