Stop finding out about a grant after it closed.
Grant staff at small and mid-sized nonprofits lose real hours every week checking Grants.gov, SAM.gov, and a rotating list of foundation and corporate giving pages by hand — and still miss opportunities that were never on anyone’s bookmark list. Discovery runs eight research lanes against those sources at once and hands you one deduplicated feed instead of a dozen browser tabs.
The actual Research page
Not a mockup — this is the live Research Command Center, captured from a running instance of the application on September 8, 2026.
Watch a sweep run
Each step below mirrors a real stage in the orchestration code (src/lib/agents/research/orchestrator.ts).
How it actually works
Discovery is not one agent — it is eight research lanes launched together. Four base families (Government Grants, Corporate Giving, Foundation Grants, Local Sponsorship) sweep broadly; four specialized passes reuse the Government and Foundation agent classes under a narrower focus (the Grants.gov API directly, state agencies, foundation directories, and faith-based sources) so each source type gets its own dedicated search rather than one generic query trying to cover all of them.
All eight lanes run concurrently via Promise.allSettled — one lane timing out or erroring never blocks the other seven, and every lane’s status is reported back so staff can see exactly which sources returned results and which didn’t on a given run.
Running lanes in parallel creates a real side effect: two lanes can independently discover the same opportunity before either has committed its row. A cross-lane deduplication pass runs after every sweep, matching first on URL and then on funder name plus program title, and keeps the earliest-created row when it finds a collision.
Every run is triggered either on demand from the Research page (per-lane buttons, or “Run All”) or on a schedule tied to an organization’s active search profiles, and is logged to an agent run history with status, items found, items processed, duration, and any error — visible from the same page.
Inputs and outputs
- Staff-configured search profile: keywords, funder categories, geography
- Focus-area weights and populations-served tags, woven into query building
- Negative filters: excluded categories and excluded funder names
- Per-agent enable/disable toggles on each profile
- Live source responses: Grants.gov, SAM.gov, state portals, foundation directories, corporate and local sponsorship pages
- Deduplicated opportunity rows: name, source, category, amount range, deadline
- A consensus validation badge on opportunities checked by both AI providers
- An agent run log per lane: status, items found/processed, duration, errors
- A count of cross-lane duplicates removed in that sweep
Where a person is in the loop
Discovery does not decide what your organization pursues. It builds a reviewed feed; a staff member decides which listing moves into eligibility scoring and, from there, into drafting. No opportunity that Discovery surfaces is added to the pipeline, scored, or drafted without that action.
The one place Discovery itself makes an automated judgment is consensus validation, and it is built to defer rather than assert: two independent AI providers each judge a finding’s existence, eligibility coherence, deadline plausibility, and amount sanity from the finding’s own text — they are instructed to return “unverifiable” rather than invent an answer, and a Verified badge only appears when both providers agree. When they disagree, the finding is left unmarked for a person to judge instead of the system picking a side.
Search profiles themselves are staff-authored and staff-maintained: which sources run, which categories count, which funders are excluded, and whether a given research family runs at all are all settings a person sets, not something the system infers on its own.
Quantified results
We don’t have a published win-rate, time-saved, or dollar-figure number for Discovery yet. Those figures depend on real pilot usage over a full grant cycle, and we’d rather show you an honest “pending” than a number we backed into. This section will be replaced with real pilot data once we have it.
What it actually connects to
Polled directly by Discovery’s research lanes.
- Grants.gov — public search API, no authentication required
- SAM.gov — opportunities search API, API-key gated
Not part of Discovery’s search — connected once at the organization level and used downstream, for example to track submission confirmations after AutoApply.
- Gmail — OAuth-connected mailbox sync
- Zoho Mail — OAuth-connected mailbox, alternate to Gmail
Security and reliability
Every research query is scoped to your organization, every sweep is logged to an auditable run history, and consensus validation is explicit about what it can and can’t verify rather than guessing. For the platform-wide security and governance model — including how approval gates work elsewhere in the platform — see the Trust and Governance page.
Customer story
COMING SOON
We don’t have a published customer story for Discovery yet. Once an organization has run it through a real grant cycle and agreed to be named, their story will go here — not a composite or hypothetical example.
FAQ
Run the free Funding Potential Scan first — a short, no-signup check of the kind of funding your organization is likely to qualify for.
Find My Best Fit Opportunities
Set up a search profile and let eight research lanes start checking government, foundation, and corporate sources for you.