> For the complete documentation index, see [llms.txt](https://assay.gitbook.io/assay-docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://assay.gitbook.io/assay-docs/resources/roadmap.md). # Roadmap What runs today, what changed since the Metropolis submission, and what comes next. **Where:** status comes from the repo README, `docs/deployments.md` and the evidence in `docs/evidence/`. Updated 23 Oct 2026. This page separates what runs today from what is next. Items move up when they ship, with a link to the evidence. ## Live on Monad mainnet | Item | Evidence | | ------------------------------------------------------------------------------------------------------------------------------------------ | ----------------------------------------------------------------------------- | | `ReceiptAnchor`, `VerifierRegistry`, `CreAttestor` v2 and `AssayAccount`, verified on Sourcify | [Network and contracts](/assay-docs/getting-started/network-and-contracts.md) | | Host 10278 (Gemma 4 31B through Google's API) and host 10316 (Kimi K2.6 through Moonshot's endpoint), signing and anchoring every response | `/health` on each host | | Passkey co-signatures (`cosign`) and per-app secp256k1 co-signatures (`cosignK`), with gas paid by the host's relayer through EIP-7702 | [Passkey co-signatures](/assay-docs/how-it-works/cosign.md) | | Grades from verifier 10279, each with an evidence bundle whose sha256 is onchain | [Verifiers and grades](/assay-docs/how-it-works/verifiers-and-grades.md) | | CRE re-checks of grades from their evidence, `agree=true` on every one so far (simulator runs relayed by our key, not a DON) | [Chainlink CRE](/assay-docs/integrations/chainlink-cre.md) | | Two paid jobs through Kanmani's escrow, settled on receipts from host 10278 | `docs/deployments.md` | | Envio indexer over mainnet and testnet in one deployment | [Indexer and GraphQL](/assay-docs/for-developers/indexer.md) | ## Since the Metropolis submission (13 Oct) | Item | Evidence | | ---------------------------------------------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------- | | Grading runs on a schedule: both hosts are re-graded every Monday and Thursday, and an alert fires if any grade is more than 5 days old (grades expire after 7) | `harness/vm_grader.sh` | | Each evidence bundle carries a `provenance.json`: the grader's git commit, who ran it, and whether they vouch for the method or only ran the code | `harness/export_grade.py` | | SDK `assay-receipts` 0.2.0 on npm. `wrap()` can check a host's grade before sending, and `chainGradeCheck` reads it from the registry instead of asking the host | [SDK reference](/assay-docs/for-developers/sdk.md) | | Kimi K2.6 hosts on OpenRouter re-graded on 20 Oct: 15 passed 32/32, one was rate-limited and got no grade | [grades\_20261020T135810Z.json](https://github.com/trudransh/Assay/blob/dev/docs/evidence/grades_20261020T135810Z.json) (on `dev` until 3 Nov) | | `max_tokens` test on every Kimi K2.6 host: the hosts that ignored the cap were sent their logs on 18 Oct, and the names and raw files are published on 25 Oct | [max\_tokens evidence](https://github.com/trudransh/Assay/blob/main/docs/evidence/max-tokens-kimi-k2.6.md) | ## Built, going live after 3 Nov These are merged on the `dev` branch and wait until Metropolis judging ends. | Item | What it changes | | -------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------- | | Host storage in SQLite | Indexed lookups, no log replayed on every restart | | Public grade feed (`harness/feed.py` on `dev`) and SDK `feedStatuses` / `toOpenRouterIgnore` | A router or agent can skip hosts that fail, or whose grade expired, with one JSON file | | `ReceiptError` in SDK 0.3.0 | A response without a usable receipt still hands you the answer, so you decide what to pay | | Check suite v1 | Adds a `max_tokens` billing check and a stop-sequence check to the grade | | Receipts count thinking tokens | `tokensOut` includes reasoning that Google's API reports only in `total_tokens` | | Uptime and low-gas alerts | Hosts, the indexer and the wallets that pay gas are checked every 15 minutes | ## Next | Item | Why | | -------------------------------------------------------------------------- | ------------------------------------------------------------------------- | | A signing gateway in front of any OpenAI-compatible router, with streaming | Receipts for traffic to hosts that don't run Assay's proxy | | A second EVM chain with the RIP-7212 P-256 precompile | The same receipts and grades readable on another chain | | x402 pay-after-check | An agent reads the grade onchain and pays per request only when it passes | | A grade from an outside verifier, re-checked by CRE | Today every grade comes from Assay's own verifier | | A real domain for the hosts and the app | Replaces the sslip.io hostname | | Monthly host report with every raw log | One place to compare every host for the models we cover | ## Later | Item | Why | | ------------------------------------------ | -------------------------------------------------------------------------------------------------------------- | | Hosts that sign from inside a TEE | A TEE quote that includes a model hash is the path to proving which weights ran | | `anchorK` for secp256k1 hosts | Hosts whose node key is secp256k1 can anchor without a P-256 key | | Keys in Cloud KMS, with a rotation runbook | The host key no longer lives in a file on the VM | | CRE live spot checks | Nodes call a host directly on a schedule, so the DON becomes a verifier and not only a re-checker | | ES384 support | NVIDIA attestation tokens use ES384. Monad has no P-384 precompile, so this needs offchain checks or a wrapper | ## Where else this shows up | Place | What it covers | | ---------------------------------------------------- | --------------------------------------- | | [Threat model](/assay-docs/security/threat-model.md) | Known limits that the roadmap addresses | | [Welcome to Assay](/assay-docs/welcome/readme.md) | Back to the start | Next: [Welcome to Assay](/assay-docs/welcome/readme.md) --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation by asking a question. Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters: ``` GET https://assay.gitbook.io/assay-docs/resources/roadmap.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.