June was a month of three big pushes. We shipped the first full version of our donor advisor platform for nonprofit research, ran end-to-end support for Filecoin's ProPGF Batch 3 round (47 shipped updates for that program alone), and published v1.0.0 of an open standard for funding opportunity data. Plus the usual pile of fixes and polish underneath. Here's the rundown.
By the numbers
- 204 PRs merged
- 115 issues shipped to production
- 15+ new features and capabilities
- 10+ UX improvements
- 30+ bug fixes
- 47 Filecoin-focused updates supporting the ProPGF Batch 3 round
A research platform for donor advisors
The biggest story of the month: we built out the donor advisor product, a research platform that helps advisors match their donors with the right nonprofits. It starts with data. We loaded IRS Form 990 and 990-PF filings going back to 2016, so every foundation and nonprofit in the system carries real financials: income, expenses, and assets for the last three years, right in the report.
On top of that data we shipped the advisor workflow. Advisors describe their donor once, and an LLM refines that into a donor persona: structured fields for org maturity, geography, faith stance, gift size, and advocacy stance, each one confirmable and editable. That persona then drives everything downstream: it pre-fills filters and weights when you create a research report, and it shows up as its own section in the synthesis. Advisors can also customize scoring weights directly, so a donor who cares more about financial health than growth gets rankings that reflect it.
Reports are built to be shared. Advisors send donors a shareable report that looks exactly like their own view, donors and advisors can leave comments on it, and when a donor wants to go deeper on a nonprofit, they can request diligence or an intro with one click. We also added deep research on individual nonprofits, including social channel activity as a signal. The whole loop, from persona to research to shared report to intro, now runs end to end.
Tool to research philanthropic funders
On top of the 990 data, we also built a tool to help nonprofits research the foundations who fund nonprofits. There are many tools in the market to do this research but surprisingly none of them allow you to do complex natural language queries. We are already seeing organic growth of this tool. We will continue to improve this tool.
Powering Filecoin ProPGF Batch 3
June was launch month for Filecoin's ProPGF Batch 3, and we supported the round end to end on filpgf.io: 47 shipped updates covering the application window opening, the timeline and content updates through the round, and the application close.
A lot of that work went into the AI assistant that applicants, reviewers, and grantees use. We tuned prompt suggestions to adapt to who's signed in, fixed response quality issues (project names instead of IDs, accurate pending-review lists, correct citations), tightened the knowledge base so program docs actually get picked up, and made sure AI evaluations always use the latest application data. Running an AI interface inside a live grant round with a demanding ops team is the best quality loop we could ask for.
We also shipped real workflow upgrades for the program: a redesigned reviewer inbox, bi-weekly snapshot reports of everything pending review (by staff, reviewer, pod, or grantee), invoice age tracking in check reports, and required Funding Requested and Completion Criteria fields on milestones so applications come in complete instead of half-filled.
An open standard for funding opportunity data
Funding opportunity data is scattered and inconsistent across every ecosystem. In June we published v1.0.0 of an open schema for funding opportunities, validated against 20+ real entries from the Karma Funding Map before locking it. The standard is public: CC0-licensed schema, MIT-licensed repo, with contribution guidelines and governance docs.
We didn't start from scratch. The schema ships with a crosswalk to DAOstar's DAOIP-5 Grants Metadata Standard and schema.org/Grant, so it plugs into the standards work the ecosystem has already done.
And it's not just a spec on paper. We shipped the core of a public read API: opportunity endpoints with full filter, search, sort, and pagination support, loaded with a 100+ entry seed dataset where every entry validates against the schema. Full public docs land in early July.
More highlights this month
- On-chain milestone completions: milestone completion state now lives fully on-chain as EAS attestations, one verifiable source of truth instead of two parallel records.
- Admins can complete milestones for grantees: when a grantee goes inactive, admins record the completion directly from the Milestones Review workspace instead of being stuck.
- Milestone management on project pages: community admins can now complete and update milestones right from project pages.
- Team member invites: community admins can add team members, and accepted members get admin access by default.
- Agentic knowledge base: the KB can now register whole sites as skill manifests, so agents can use data portals and query interfaces instead of just reading chunked text.
- Trusted agent identity: API calls made by our MCP agent now carry the verified caller identity end to end, hardening authorization across the platform.
- Search our funder database from ChatGPT or Claude: we published instructions for querying Karma's nonprofit funding data straight from your AI assistant.
UX polish
- Redesigned the reviewer inbox page.
- Cleaned up the funding applicant dashboard.
- Foundation, grant, and nonprofit URLs now use clean name slugs instead of UUIDs.
- Monthly reports now default to holistic graphs, with per-project breakdown a toggle away.
- AI interface prompt suggestions adapt to whether you're signed in and who you are.
Performance and reliability
- Fixed an out-of-memory issue in the indexer.
- The in-product chatbot now degrades gracefully on heavy turns and continues reliably instead of stalling silently.
- Killed a request storm on project pages that fired hundreds of duplicate fetches.
- Ported legacy V1 write endpoints to V2 with proper server-side authorization, closing a real security gap.
- Our Sentry-to-PR automation now spins up an agent to fix new production errors and open a PR automatically.
Notable fixes
- Editing a milestone no longer creates a duplicate, it updates the original.
- Embedded wallets (email and Google login) now confirm chain switches before signing, and Google login no longer creates duplicate wallets.
- Fixed a nonce collision that broke gasless milestone deletion.
- Community admins can now download invoices (previously blocked by an owner-only check).
- Admins and reviewers can submit to closed funding programs as intended, apply surfaces now respect roles.
- Fixed application submission errors that could wipe an application at login.
- 10+ Sentry-tracked production fixes across the indexer and frontend.
What's next
July picks up right where June left off: public API docs for the funding-data standard are already live at /v1/docs, and there's more coming on the donor advisor side. As always, huge thanks to our dev team for the hard work behind all of this. Got feedback or feature requests? Join our Telegram group and let us know.