# Fix list: GAIA

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/gaia, the October 2026 research run, assessed 12 October 2026. Grade D, 49.5 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 GAIA: 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, 20 out of 100, up to 10 more on the total

Why it scored 20: Scored on the hosted service. No payment protocol (0). The pricing page shows one plan, Pro at $30 a month, with usage limits it does not state (10). A no-card free tier takes half marks, because the README and docs describe a Free plan while the pricing page says GAIA is a paid product, and we did not sign up to see which holds (10). Access needs a person to sign up in a browser, and self-hosting needs a WorkOS account and model keys (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. Security & auth, 54 out of 100, up to 8.1 more on the total

Why it scored 54: Sign-in runs through WorkOS and integrations connect by OAuth on the provider's consent screen, with custom MCP tokens Fernet-encrypted in PostgreSQL and disconnect revoking access. Gmail takes the single broad `gmail.modify` scope and Composio, a third party, holds the built-in connections (22 of 30). Three approval modes, per-tool overrides and forced approval for adding an MCP server, pairing a device or disconnecting an account, but the source default is `always_allow`, commented as nothing gated (9 of 20). In Auto mode the judge trusts only the user's quoted words, fences untrusted text with a random marker and fails towards asking. That protection is off in the default mode and no guidance for owners was found (9 of 15). The chat shows each tool call and approvals are kept in a ledger, while audit logs are listed for Enterprise only (7 of 15). SECURITY.md promises a reply in 48 hours and coordinated disclosure, and the repository runs gitleaks and semgrep. No security.txt, bounty, certification or published advisory was found (7 of 20).

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.

## 3. Schema & documentation, 51 out of 100, up to 8 more on the total

Why it scored 51: GAIA has no public API, so this is read on what a person or an agent can learn about the product from its docs and source. In production the backend turns off its OpenAPI document, and the generated schema sits only in the repository (5 of 25). llms.txt on heygaia.io and docs.heygaia.io, with Markdown copies of every docs page (10). The guides say what each feature is for, and ARCHITECTURE.md maps every service and file, but no page lists which actions run without approval (12 of 20). The custom MCP form and the workflow builder have typed fields, and tools are otherwise chosen by the assistant (6 of 15). Guides carry examples and no page documents errors (7 of 15). Dated release notes and a tag for each component, though the notes stop at 26 August while tags exist for 13 and 14 September (11 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.

## 4. Agent ergonomics, 54 out of 100, up to 7.5 more on the total

Why it scored 54: Read on what the owner has to supply and manage. A comms agent delegates to an executor and to one subagent per integration, each scoped to its own tools, and tools are found by retrieval instead of loading every definition (18 of 25). Usage caps and notification settings exist, with no control over reply length found (8 of 20). Limit messages name the limit and its reset time, and other errors are not documented (8 of 20). An approval ledger records decisions, declined calls are remembered, and MCP `destructiveHint` is honoured as an escalation only (12 of 20). Connecting an account starts jobs with no setup and there is a CLI, but there is no SDK and the default approval mode runs everything (8 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.

## 5. Reliability, 67 out of 100, up to 6.6 more on the total

Why it scored 67: Graded as the hosted service at heygaia.io, which the README recommends, with the hosted lines. A BetterStack status page lists the backend, the web app and four bots with a year of history (20). It lists no incident in the 90 days to 12 October 2026. The last entries are a Discord bot outage of 6 to 9 May and maintenance of 16 to 19 April, and we did not open the previous-incidents list (25). The docs give daily and monthly caps for Free and Pro, while the 26 August notes say Pro runs on cost-based limits with no figure (10). The docs say a limit is reported in the conversation with its reset time, and no retry guidance was found (5). An SLA is named only as part of Enterprise, with no terms published (0). The web app and bots carry no beta label, the desktop and mobile apps do, and every component is below 1.0 (7).

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.

## 6. Transparency & trust, 60 out of 100, up to 3.5 more on the total

Made of editorial 50, provenance 70.

Why it scored 60: The full source is public under PolyForm Noncommercial 1.0.0, which is not an OSI licence and bars business use without an enterprise licence (20 of 30). The privacy policy of 7 August 2026 gives retention periods, Google Limited Use terms and a 72-hour breach notice, but it conflicts with the pricing page FAQ on sharing with model providers, allows submitted content to be used to improve the service, and gives a contact address on another domain. No DPA was found (14 of 30). The terms promise 30 days' notice before the service or a paid feature is discontinued, and no deprecation policy for single integrations or tools was found (10 of 20). The policy names categories of processor and no company or data location. The source and docs name OpenRouter, Google, Composio, WorkOS, E2B, Dodo Payments, PostHog and Sentry as integrations (6 of 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: heygaia.io, no registry record we could read (0 of 15)
- Terms of service: read, states 6 of the 7 things a reader expects, and has 2 clauses that cost points (5.1 of 10)
- security.txt: not found (0 of 10)

## 7. Maintenance & community, 78 out of 100, up to 1.9 more on the total

Why it scored 78: The last tagged releases are `api-v0.24.0`, `web-v0.25.0` and `desktop-v0.7.1` of 14 September 2026, 28 days before this check (30). Three sets of tags in 90 days, on 26 August, 13 September and 14 September (20). 91 commits on master in the 30 days to 12 October. The 40 open items are mostly the team's own pull requests, the few open bug reports each have one comment, and the site names Discord, WhatsApp and email support with a 24 to 48 hour reply claim we did not test (15 of 25). No SDK and no MCP registry entry. `@heygaia/cli` 0.6.0 on npm matches the `cli-v0.6.0` tag (8 of 15). Dependabot and lockfiles are in place, but Quality Checks, Release Please and Deploy Web failed on the latest push to master on 12 October (5 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.

- -3. The release notes of 26 August 2026 say GAIA pauses and asks before a tool that deletes, sends or overwrites, the privacy policy (7 August 2026) says destructive actions require explicit approval, and the pricing page says anything outward-facing waits for the owner's OK. The source at `api-v0.24.0` (14 September 2026) and on master sets the default mode to `always_allow`, commented as nothing gated until flipped after launch. Read from source, not from a live account. https://github.com/theexperiencecompany/gaia/blob/master/apps/api/app/models/hil_models.py
- -2. The pricing page FAQ, read on 12 October 2026, says GAIA never shares data with model providers. Section 3.1 of the privacy policy says content, files and data from connected accounts are shared with third-party AI model providers. https://heygaia.io/pricing and https://heygaia.io/privacy

## 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.

- Kind is `platform`, the kind the brief gives a web, desktop or phone app, because the hosted assistant has no public API. It was graded as a hosted service with `metrics.kind` remote, since the README recommends the cloud and prices are published for it. The self-hosted route is recorded as facts and not scored.
- The default approval mode was read from source at `api-v0.24.0` and master. Whether the hosted service sets another default for new accounts was not checked, because we did not create an account.
- Whether a new hosted account can still use a Free plan. The README and docs say yes and the pricing page says GAIA is a paid product.
- The Terms of Service forbid using automated systems, including robots, spiders or data mining tools, to access the Service. Recorded as a fact with no deduction. It matters before any probe is run.
- unchecked: the status page's previous-incidents list, so the 90-day record rests on the main status page.
- unchecked: the domain registration date for heygaia.io.
- unchecked: which companies process data for the hosted service and where. The privacy policy names none, and the names here come from the source and self-hosting docs.
- unchecked: how account deletion works in the app. The privacy policy routes deletion requests through email, and it gives a heygaia.so address.
- unchecked: reply times on Discord, WhatsApp and email support.
- The docs guides were read from the repository's docs folder, not from docs.heygaia.io pages. Only llms.txt and robots.txt were fetched from that host.

## Weaknesses

- The default approval mode in the source is `always_allow`, commented as nothing gated, while the privacy policy and the 26 August release notes say destructive actions wait for approval.
- The pricing page FAQ says data is never shared with model providers. The privacy policy says content, including data from connected accounts, is shared with them.
- The README lists a Free plan at $0 forever, while the pricing page shows only Pro and says GAIA is a paid product.
- Gmail connects with the `gmail.modify` scope, and built-in integrations run through Composio, so a third party holds those OAuth connections.
- No security.txt, bug bounty, certification, subprocessor list or DPA was found, and self-hosting needs a WorkOS account and five data services.
- The Terms of Service forbid automated systems accessing the Service. Recorded as a fact, and it matters before any probe is run.

## 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.

- Set Settings > Approvals to Always ask or Auto before connecting mail. The default in the source runs every action, including send and delete, without asking.
- Expect no public API on the hosted service. The backend turns off its OpenAPI document in production, so reach GAIA through its apps, bots or CLI.
- Check the plan before relying on the hosted service. The README and docs describe a Free plan, and the pricing page says GAIA is a paid product.
- For self-hosting, prepare Docker, Node.js 20, a WorkOS account and a model key first, and confirm the use is noncommercial under the licence.
- Add outside tools as MCP servers by URL on the Integrations page. GAIA asks before adding a server in every approval mode.

## 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.