Vespa

# Fix list: Vespa

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/vespa, the October 2026 research run, assessed 9 October 2026. Grade BB, 70 out of 100.

This is everything the published grade says the listing lacks, the biggest possible gain to the total first. It comes from the reason given for each score, the checklist each category was scored against (https://www.anchorterminal.com/benchmark/#checklist), the provenance checks, the deductions, what we couldn't check and what the review panel asked for. A fix counts at the next check, once it's public.

For a coding agent working on Vespa: work through the items below in the product, its docs and its public pages. Each category gives the reason for its score, with the points each checklist item earned, and the checklist itself, so the gap is the items that earned less than their points. Change the product, not the wording, and keep a note of what you changed and where it's published.

## 1. Payments & pricing, 40 out of 100, up to 7.5 more on the total

Why it scored 40: Scored on Vespa Cloud, the paid hosted option. No x402, MPP or L402 (0). Unit prices per vCPU-hour, GB-hour of memory and GB-hour of disk are public for four plans without a login (20). The trial has $300 of credits and needs no card (20). A person signs up in the browser and logs in through Auth0 before any deployment (0).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-payments):

The published rubric, also on the [x402 page](https://www.anchorterminal.com/x402/).

- 40, a machine payment protocol (x402, MPP or L402) on the tool's own endpoints. 10 to 30 when it covers only some endpoints or only goes through a third party, and the note says which.
- 20, per-call or per-unit pricing published without a login. 10 for public plan-only pricing, 0 for "contact sales" or prices behind a login.
- 20, a free tier or trial that doesn't need a card.
- 20, autonomous onboarding, meaning an agent can get access without a person signing up in a browser (keyless use, x402, a programmatic key API).

Payment platforms and agent wallets rarely charge for their own API over a machine protocol, so the first line has steps for them, and the highest one that applies counts. 40 when x402, MPP or L402 runs on all their own endpoints, 30 when it runs on part of their own API, 25 when their merchants can accept one, 20 for running a facilitator, 15 for paying as a buyer, and 0 when the only protocol is their own. Merchant acceptance sits above a facilitator because the platform's own customers can charge agents through it, while a facilitator settles for sellers who wire up the protocol themselves. The counter-argument (a facilitator does more for the protocol as a whole) has a point. Each note says which step applied.

Open-source software you run yourself is scored on its hosted or paid option if it has one. A free, self-hosted package with nothing to buy gets 20, 20 and 20 for the last three lines, and 0 to 40 for the first only if it ships a payment protocol.

## 2. Schema & documentation, 58 out of 100, up to 6.8 more on the total

Why it scored 58: No OpenAPI file or other machine-readable contract was found in the docs index of 384 pages (0). `llms.txt`, `llms-full.txt` and a Markdown twin of every page at `.html.md` (10). The query and `/document/v1` reference pages describe each parameter, its type and default, and the document API guide says when to use conditional writes and visiting (15). Parameters are typed in tables, and documents are validated against the deployed schema, but queries travel as a YQL string and the optional MCP tool takes a free-form `parameters` object (10). Examples on most pages and status code tables for both APIs (13). Semantic versioning with release notes for each major version and a commit list per release in Vespa Factory, but no written changelog per release was found (10 of 15).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-schema):

APIs and MCP servers.

- 25, a machine-readable contract (a public OpenAPI file or similar; for MCP, typed JSON Schema inputs on every tool).
- 10, llms.txt or Markdown docs served for agents.
- 0 to 20, descriptions that say what a tool is for, when to use it and when not to, read from the tool definitions in the source or the API reference.
- 0 to 15, typed inputs with enums, constraints and required fields, and no free-form JSON blobs.
- 0 to 15, examples and documented error responses.
- 15, versioning and a public changelog.

Models are read from the API reference, the OpenAPI file, llms.txt, the structured-output and tool-use docs and the model cards. Frameworks from docs a model can follow, typed interfaces, examples and the API reference.

## 3. Security & auth, 75 out of 100, up to 4.4 more on the total

Why it scored 75: Data-plane clients authenticate with mTLS certificates or bearer tokens, each limited to `read`, `write` or both in `services.xml`. Tokens have versions for rotation, expire after 30 days by default and can be revoked, and are sent in a header (30). Read-only clients are supported. No confirmation step for deletes was found in the reviewed pages (13). Queries return stored documents as written, and no prompt-injection guidance was found (5). Control-plane operations go to an audit log that is available on request from support, and data-plane access logs are JSON, kept for at most 30 days (10 of 15). A valid security.txt, a disclosure programme on Intigriti with no public bounty, a SOC 2 attestation per the security guide, and one advisory published on GitHub in September 2026 (17).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-security):

- 0 to 30, the credential model. 30 for OAuth 2.1 with scopes, or scoped and revocable keys with rotation. 20 for plain revocable API keys. 10 for one all-powerful key. 10 off when a secret can travel in a URL query string as a documented option.
- 0 to 20, read-only or least-privilege modes, and confirmation or approval for destructive actions.
- 0 to 15, prompt-injection posture where the tool returns untrusted content (documented mitigations or guidance). A tool that returns no untrusted content gets 10.
- 0 to 15, audit logs or per-call visibility for the operator.
- 0 to 20, a security programme. security.txt or a disclosure policy, a bug bounty, SOC 2 or ISO 27001, advisories handled in public.

Models are read for retention, whether API data trains models (and whether that's off by default), zero-retention options and certifications. Frameworks for telemetry defaults, approval hooks, guardrails and sandboxing.

## 4. Reliability, 80 out of 100, up to 4 more on the total

Why it scored 80: Graded as a hosted service, on Vespa Cloud. Instatus status page at status.vespa.ai with components for the console, infrastructure, enclave and every zone, and a history feed (20). In the 90 days to 9 October 2026 the feed shows one incident, failing staging tests in AWS us-east-1c for 7 hours 27 minutes on 19 August, scoped to test and staging zones, with production zones at 100.0% on the page. Minor only (20). No request rate limit is published. The tenant quota is $2 an hour on trial and $10 an hour on other plans, the query timeout defaults to 0.5s and `hits` is capped at 400, so limits with numbers but none on request rate (7 of 15). `/document/v1` returns 429 when too many feed operations are in flight, the HTTP best practices page asks for capped exponential backoff with jitter and no retry on client errors, and writes by document ID with a `condition` are safe to repeat. No `Retry-After` header is documented (13 of 15). SLA of 99.5% monthly uptime for paying customers (10). Generally available (10).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-reliability):

Hosted APIs, MCP servers, models and platforms.

- 20, a public status page with component history (Statuspage, Instatus, BetterStack or the vendor's own).
- 0 to 30, the incident record for the last 90 days on that page. 30 for a clean record or trivial incidents only, 20 for minor incidents only, 10 for one major outage (an hour or more of a core API down, or errors across the board), 0 for several. 5 when there's no history we could read, and the note says so.
- 15, rate limits documented with numbers.
- 15, documented 429 or overload handling (Retry-After, backoff guidance), and idempotency keys or safe-retry guidance where writes are involved.
- 10, an SLA published for any paid tier.
- 10, the surface agents use is generally available, not beta or preview.

Local packages, SDKs, frameworks and stdio MCP servers.

- 20, installs from an official package with supported runtimes stated.
- 25, a public CI and test suite, passing on the default branch.
- 0 to 25, open crash or regression issues relative to activity (25 for few and handled, 0 for many, old and unanswered).
- 15, semver discipline and breaking changes called out in a changelog.
- 15, version 1.0 or later, or declared stable.

Protocols are read from their reference implementations, the public facilitators or servers, spec stability and test vectors.

## 5. Agent ergonomics, 80 out of 100, up to 3.3 more on the total

Why it scored 80: Responses can be sized with `hits`, `offset`, `presentation.summary` document summaries and `fieldSet` on document reads (20). Paging by `hits` and `offset`, continuation tokens when visiting, and structured filters in YQL (20). Both APIs document their status codes, and errors carry a `message` field, but the query API's errors are Java error codes mapped to HTTP codes (15). Puts by document ID repeat safely, `condition` gives test-and-set with 412 on failure, and `dryRun` exists on document writes. No idempotency key (15). Official clients in Python and Java plus the Go CLI, but nothing answers until an application package with a schema is deployed, and the default credential is a client certificate (10 of 15).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-ergonomics):

- 0 to 25, context cost. For MCP, the number and size of the tool definitions (25 for ten or fewer compact tools, 15 for 11 to 30, 5 for more than 30, plus up to 10 back for toolsets, dynamic loading or read-only subsets). For APIs, whether responses can be sized (field selection, limits, summaries).
- 20, pagination, filtering and output-size controls.
- 20, actionable, documented error responses, codes and messages an agent can recover from.
- 20, idempotency or safe retries, and for MCP the `readOnlyHint` and `destructiveHint` annotations.
- 15, sensible defaults, few required parameters, and official SDKs in at least two languages.

Models are read for tool use, structured output, prompt caching, context length, batch and SDKs. Frameworks for how much code and how many defaults a tool-calling agent with MCP needs.

## 6. Transparency & trust, 85 out of 100, up to 1.3 more on the total

Made of editorial 86, provenance 84.

Why it scored 85: Apache-2.0 for the engine and clients (30). The privacy policy, a DPA with 24-hour breach notice and a security white paper with a 30-day log limit are public. The DPA dates from 18 October 2023, names Vespa.AI Norway AS where the privacy policy names Vespa.ai AS, and refers to a Service Agreement not found on the site. The privacy policy states no retention period (22). Deprecated APIs and settings are removed only at a major version, with compile and deployment warnings before that, and the Vespa 9 notes list planned removals without dates (14). Sub-processors are listed with data, purpose and location, updated 12 August 2026, and logs stay in the application's region (20).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-transparency):

- 0 to 30, source availability and licence clarity. 30 for open source under an OSI licence, 15 for closed with clear terms, 0 for unclear terms.
- 0 to 30, data handling and retention statements that agree with each other (privacy policy, DPA, retention periods, subprocessors).
- 0 to 20, a deprecation policy or notices with dates.
- 0 to 20, telemetry disclosed with an opt-out (local software), or subprocessors and data locations disclosed (hosted).

The other half of Transparency and trust is the provenance score, computed from checked facts (below). The category score is the mean of the two.

Provenance checks not met in full (half of this category, computed from checked facts):

- Domain age: vespa.ai, registered 2017-12-16 (8 years) (11 of 15)
- Terms of service: not found (0 of 10)
- Privacy policy: read, states 5 of the 8 things a reader expects (7.8 of 10)

## 7. Maintenance & community, 91 out of 100, up to 0.8 more on the total

Why it scored 91: Tag v8.763.13 is dated 29 September 2026 and pyvespa 1.2.8 reached PyPI on 1 October 2026, both within 30 days (30). 16 release tags between 12 July and 29 September 2026 (20). The repository showed 225 open issues and 42 open pull requests, the master branch took merges on 8 October, and recent issues carry replies, though we did not read who replied or how fast (18 of 25). pyvespa, the Java feed client and the CLI are current (15). `renovate.json` and 11 GitHub workflows are in the repository. The main build runs on Buildkite and its state was not read (8 of 10).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-maintenance):

- 0 to 30, time since the last release, or the last published model or API change for a closed service. 30 within 30 days, 20 within 90, 10 within 180, 0 older.
- 20, at least three releases or dated changelog entries in the last 90 days.
- 0 to 25, responsiveness. Issues and pull requests answered on GitHub (the open issues and how recent the replies are). For closed services, a public changelog and a support or community channel that answers, 0 to 15.
- 15, presence in the official MCP registry under a verified namespace (MCP servers), or current official SDKs (APIs and models).
- 10, package health, current dependencies and CI.

Models are read for deprecation notice periods and model churn rather than release counts.

## Deductions

Each comes off the total. A fixed and documented problem counts for less at the next check.

- GHSA-v86h-xr9p-pgm5, moderate severity (CVSS 6.8), published 4 September 2026. Two local file inclusion flaws in the deployment API let an authorised deployer read files readable by the config server process, in versions up to 8.746.22. Patched in 8.748.3 and disclosed in public, so a small deduction (https://github.com/vespa-engine/vespa/security/advisories/GHSA-v86h-xr9p-pgm5).

## What we couldn't check

What we couldn't read counted as absent. Publishing it on a page a plain HTTP fetch can read (not only in a browser) lets the next check count it.

- unchecked: the trust centre at trust.vespa.ai, which is drawn by script, so the SOC 2 attestation rests on the security guide's statement
- unchecked: the Vespa Service Agreement or terms of use, which the DPA refers to and which was not linked from the product terms page
- unchecked: exact GitHub star count and issue reply times, because the GitHub API answered with a rate limit. The 7,100 stars figure is the rounded count on the repository page
- unchecked: whether the main Buildkite build passes on the default branch
- unchecked: whether Vespa Cloud was affected by GHSA-v86h-xr9p-pgm5, which the advisory does not say
- Whether self-hosted Vespa sends any usage telemetry. No statement was found in the pages reviewed
- Whether a written changelog per release exists outside the commit lists in Vespa Factory
- The lead named a Java feed client and no MCP server. Both hold, with the addition that the engine ships an optional MCP component for a user's own application

## Weaknesses

- No OpenAPI file was found in the docs index. The APIs are described in reference pages only
- Vespa.ai states it has no official public MCP server. An MCP endpoint must be added to the user's own application
- No call works until a schema and `services.xml` are deployed as an application package
- The Service Agreement named in the DPA is not on the product terms page, which links only the privacy policy, DPA and SLA
- Advisory GHSA-v86h-xr9p-pgm5 (moderate, 4 September 2026) let authorised deployers read config server files before 8.748.3

## What costs an agent a turn today

The notes we give agents before they call it. Each one is a workaround an agent shouldn't need.

- Deploy an application package with a schema before feeding or querying. Use `vespa deploy` and wait for the endpoint to report ready
- Send a token as `Authorization: Bearer` only to the endpoint marked Token. The mTLS endpoint needs the client certificate and key
- Set `hits`, `offset` and `presentation.summary` on queries. `hits` is capped at 400 by default and the query timeout defaults to 0.5s
- On 429 from `/document/v1` reduce the feed rate and back off. On 507 stop feeding, because the cluster is out of memory or disk
- Use the `condition` parameter for test-and-set writes and expect 412 when it fails or the document is missing

## When it's done

Send what changed and where it's published as a dispute (https://www.anchorterminal.com/builders/#disputes, or `POST https://www.anchorterminal.com/api/v1/contact` with `"kind": "dispute"`). Disputes are answered in public, and the listing is checked again by the same checklist. Paying for an audit or a listing claim changes nothing here.