Funding

No grant, sponsor, donor or pilot partner funds Garnet today. Garnet is pre-revenue, and every goal below is unfunded.

Channels that are live

None. There is no GitHub Sponsors listing, no Open Collective and no Patreon, and this site carries no payment link. A channel appears here the day it opens, and not before.

The support that exists today costs nothing: starring, filing issues and contributing to the repository. At one maintainer, a reproducible bug report is worth more than a small donation.

What funding would buy

Nine goals, priced 2026-09-15. Each range is an estimate built from published wage and vendor figures, not a quote: the low end prices labor at the BLS median for software developers, the high end at senior contractor rates. The numbering is the order they were costed, not a commitment to that order, and no total is published because the order is not fixed.

1. Reject illegal enum values at check time. Delivered 2026-09-18 without funding (D-107, D-107b): in safe functions garnet check now rejects a unit variant given a payload, a payload variant given none or the wrong count, and an undeclared variant — including through module paths, aliases and generic enums — and an impl on one enum can no longer vouch for a same-named enum elsewhere. What remains is bounded and unfunded: managed def bodies are not inspected, by documented contract. This was checker correctness; it changed no capability claim. Estimate $3,866–$17,920 (priced 2026-09-15; the check-time cure shipped, the managed-mode extension is not separately priced)
2. Authority-sized review as a GitHub Action. garnet diff-caps exits 1 on a three-character @caps() → @caps(net) widening and 0 on the unchanged control. An action routes that change to review whatever the diff size. Bound: declared surface only — it does not prove the absence of undeclared authority. Estimate $5,155–$17,920
3. Signed downloads on macOS and Windows. Release assets are checksum-verified against a GPG-signed SHA256SUMS. The binaries themselves are not code-signed, so Windows can warn on first run and Studio is build-from-source. Signing proves who published a file and that it was not altered; it says nothing about what the program does. About $219/yr of that range is recurring vendor fees. Estimate $5,374–$36,059
4. A paid independent reviewer for trust-kernel changes. Every merge rule assumes an independent reviewer, but no reporter checks reviewer lineage, so independence is a convention rather than a checked field. Money buys a named outside reviewer; code makes the field a gate check. Paid review buys no merge authority, and the release tag stays the maintainer's. Estimate $10,031–$92,960
5. Close the checker's known blind spots. The capability scope table lists where garnet check is silent: the body of an unannotated function, and calls that reach a primitive through a function value, closure or actor handler rather than a named call. (The cycle blind spot, the undeclared time/uuid rows and the capability-less memory::* natives were closed on main after the v0.8.2 release; no release carries those fixes yet.) Closing these shortens the scope table's "may not say" list. It does not make the checker complete. Estimate $7,733–$35,840
6. Installers that verify the release signature, and Linux packages for older distributions. install.sh and install.ps1 check SHA-256, but neither fetches or verifies SHA256SUMS.asc. The Linux assets need glibc 2.39 or newer, so they do not install on Debian 12, Ubuntu 22.04 or RHEL 9. Estimate $5,155–$26,880
7. An outside security review of the capability boundary. No outside team has tested the fenced claim: that the 24 entry-gated primitives trap unless the program entry declares the capability, and that the garnet CLI and the default Interpreter::new() path are deny-by-default. The review would be scoped to that claim, and its final price set after scoping. Estimate $35,155–$235,840
8. A runtime execution budget for @bounded. Only @caps and @max_depth trap today; @bounded is declared, not enforced. A deterministic step-budget trap on both backends would move that one fence. Memory and time ceilings stay declared. Estimate $10,310–$53,760
9. An MCP host that refuses undeclared tool calls. garnet mcp-serve is one sealed local host: tool capability declarations are self-declared and nothing intercepts a call. Interception plus a deny path makes the declaration checked for this host only, not for MCP in general. Estimate $7,733–$35,840

What money does not buy

• Merge authority. A paid reviewer reviews. The maintainer merges, and the release tag stays the maintainer's.
• A claim. Funding a goal does not move a fence from declared to enforced. A trap test that could have failed does, and until one exists the public copy keeps saying declared.
• Roadmap position. Money can finish work that is already the right work. It does not buy a feature onto the language, and it does not reorder the goals above.
• A support commitment. Garnet is a research-grade prototype at v0.x. There is no SLA, no long-term support branch and no compatibility promise attached to any contribution.

Where it would go

Contributions would be received by Island Development Crew, the entity that maintains Garnet, and spent on the goals above. Island Development Crew's own running costs do not come out of Garnet money. The binding constraint is engineering hours, not infrastructure.

Commercial & sponsorship inquiries

Write to hello@garnet-lang.org, or open a GitHub issue if the question suits a public thread. There is no inquiry form yet.