{
  "fixes": {
    "slug": "assemblyai-voice-agent",
    "name": "AssemblyAI Voice Agent API",
    "listing": "https://www.anchorterminal.com/tools/assemblyai-voice-agent",
    "markdown": "# Fix list: AssemblyAI Voice Agent API\n\nFrom Anchor Terminal's listing at https://www.anchorterminal.com/tools/assemblyai-voice-agent, the October 2026 research run, assessed 9 October 2026. Grade B, 64.4 out of 100.\n\nThis 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.\n\nFor a coding agent working on AssemblyAI Voice Agent API: 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.\n\n## 1. Reliability, 60 out of 100, up to 8 more on the total\n\nWhy it scored 60: Scored with the hosted lines. Statuspage at status.assemblyai.com with component history, but its components are the Asynchronous API, Streaming API and LLM Gateway and none is the Voice Agent API (15 of 20). No incident in the 90 days to 9 October 2026 names the Voice Agent API. Five name the realtime speech-to-text it runs on, including about two hours of degraded Universal-3.5 Pro streaming in us-east on 31 July and about two hours of rejected Universal-3.6 Pro Realtime sessions on the US endpoint on 6 October. Whether agent sessions were affected is not stated (15 of 30). A per-account concurrent-session limit is enforced and no figure is published for it (0 of 15). The events reference names `at_capacity`, `concurrency_exceeded` and `internal_error` as the only retryable codes and says to reconnect with backoff, sessions resume for 30 seconds, and webhooks carry a stable `event_id`. No idempotency key on `POST /v1/agents` (10 of 15). A Service Level Agreement of 99.9 per cent a month with credits is published, written against an Order Form (10). The launch note of 29 April 2026 says the API is available to all customers and no beta label was found (10).\n\nThe checklist (https://www.anchorterminal.com/benchmark/#checklist-reliability):\n\nHosted APIs, MCP servers, models and platforms.\n\n- 20, a public status page with component history (Statuspage, Instatus, BetterStack or the vendor's own).\n- 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.\n- 15, rate limits documented with numbers.\n- 15, documented 429 or overload handling (Retry-After, backoff guidance), and idempotency keys or safe-retry guidance where writes are involved.\n- 10, an SLA published for any paid tier.\n- 10, the surface agents use is generally available, not beta or preview.\n\nLocal packages, SDKs, frameworks and stdio MCP servers.\n\n- 20, installs from an official package with supported runtimes stated.\n- 25, a public CI and test suite, passing on the default branch.\n- 0 to 25, open crash or regression issues relative to activity (25 for few and handled, 0 for many, old and unanswered).\n- 15, semver discipline and breaking changes called out in a changelog.\n- 15, version 1.0 or later, or declared stable.\n\nProtocols are read from their reference implementations, the public facilitators or servers, spec stability and test vectors.\n\n## 2. Payments \u0026 pricing, 40 out of 100, up to 7.5 more on the total\n\nWhy it scored 40: No x402, MPP or L402 (0 of 40). One per-unit price on the public pricing page, $4.50 an hour billed per second (20). $50 of credit with no card, which the billing docs say covers the Voice Agent API (20). A person has to sign up in a browser to get the first key, and no key-creation API was found (0).\n\nThe checklist (https://www.anchorterminal.com/benchmark/#checklist-payments):\n\nThe published rubric, also on the [x402 page](https://www.anchorterminal.com/x402/).\n\n- 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.\n- 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.\n- 20, a free tier or trial that doesn't need a card.\n- 20, autonomous onboarding, meaning an agent can get access without a person signing up in a browser (keyless use, x402, a programmatic key API).\n\nPayment 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.\n\nOpen-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.\n\n## 3. Security \u0026 auth, 62 out of 100, up to 6.7 more on the total\n\nWhy it scored 62: Plain API keys per project, created and deleted in the dashboard, with no scopes or expiry found (20 of 30). Browser clients pass a token in the `token` query parameter. We took no deduction because the token is one-time and redeemable for at most 600 seconds, which is a judgement call. A read-only Reader role exists for dashboard members but not for keys. HTTP tool header values are write-only and encrypted at rest, tool calls go only to public `https` hosts with redirects refused, and the docs show how to gate tools by conversation state. No built-in confirmation step before a tool that writes (10 of 20). Caller speech and tool results reach a model. Tool arguments are drawn only from user turns and tool results since 26 August 2026 and tool responses are capped, but no prompt-injection guidance was found (8 of 15). Sessions keep a timeline with tool calls, webhook deliveries are logged per session, usage is broken down by key, and the dashboard has activity logs (12 of 15). The security page states a SOC 2 Type 2 report and third-party penetration tests. No bug bounty or disclosure policy was found, and robots.txt closes `/.well-known`, so security.txt went unread (12 of 20).\n\nThe checklist (https://www.anchorterminal.com/benchmark/#checklist-security):\n\n- 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.\n- 0 to 20, read-only or least-privilege modes, and confirmation or approval for destructive actions.\n- 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.\n- 0 to 15, audit logs or per-call visibility for the operator.\n- 0 to 20, a security programme. security.txt or a disclosure policy, a bug bounty, SOC 2 or ISO 27001, advisories handled in public.\n\nModels 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.\n\n## 4. Agent ergonomics, 72 out of 100, up to 4.6 more on the total\n\nWhy it scored 72: Scored as an API. Events are small typed JSON, HTTP tool responses are capped at 8 KiB before they reach the model, and the session list leaves artefacts out (18 of 25). `GET /v1/sessions` pages by `cursor` with `limit` from 1 to 200 and filters on `status` and `agent_id` (18 of 20). Errors carry `code`, `message` and sometimes `param`, and REST validation errors name the field (17 of 20). No idempotency keys on REST writes. Sessions resume within 30 seconds, webhooks dedupe on `event_id`, and `hold` mode silences the agent during long tools (10 of 20). An agent needs three fields (`name`, `system_prompt`, `voice`). The official Python and Node SDKs have no Voice Agent client, so callers write JSON over a socket or start from the two starter repositories (9 of 15).\n\nThe checklist (https://www.anchorterminal.com/benchmark/#checklist-ergonomics):\n\n- 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).\n- 20, pagination, filtering and output-size controls.\n- 20, actionable, documented error responses, codes and messages an agent can recover from.\n- 20, idempotency or safe retries, and for MCP the `readOnlyHint` and `destructiveHint` annotations.\n- 15, sensible defaults, few required parameters, and official SDKs in at least two languages.\n\nModels 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.\n\n## 5. Maintenance \u0026 community, 64 out of 100, up to 3.2 more on the total\n\nWhy it scored 64: Universal-3.6 Pro Realtime, the speech-to-text the pricing page lists inside the Voice Agent API, shipped on 29 September 2026. The last changelog entry tagged Voice Agent API is 26 August 2026. We scored on the August entry, the last one tagged for this API, 44 days before the check (20). Ten entries tagged Voice Agent API between 13 July and 26 August 2026 (20). A public changelog that records fixes, a support address and a Discord link we did not test (11 of 15 for a closed service). The Python SDK (1.6.1) and Node SDK (4.41.5) were tagged on 24 September 2026 but have no Voice Agent client. The two starter repositories were last changed on 16 August 2026 (8 of 15). The starters need no dependencies and have no licence file, tests or CI (5 of 10).\n\nThe checklist (https://www.anchorterminal.com/benchmark/#checklist-maintenance):\n\n- 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.\n- 20, at least three releases or dated changelog entries in the last 90 days.\n- 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.\n- 15, presence in the official MCP registry under a verified namespace (MCP servers), or current official SDKs (APIs and models).\n- 10, package health, current dependencies and CI.\n\nModels are read for deprecation notice periods and model churn rather than release counts.\n\n## 6. Transparency \u0026 trust, 66 out of 100, up to 3 more on the total\n\nMade of editorial 51, provenance 81.\n\nWhy it scored 66: A closed service with a clear service agreement. The starter repositories have no licence file (15 of 30). The Terms of Service grant a licence to train on Customer Data by default, the opt-out needs a paid plan, and the retention page gives periods for streaming, sync, dictation, asynchronous and LLM Gateway data but none for Voice Agent sessions, which are recorded and stored. A DPA is published and the privacy policy states Data Privacy Framework certification (14 of 30). The changelog dates deprecations ahead of time, such as the 11 July 2026 notice of a change on 7 August, and the terms promise reasonable prior notice of major changes. No written deprecation policy was found (12 of 20). The subprocessor list sits on a third-party trust centre we did not read. US and EU endpoints are mentioned, with no EU host named for the Voice Agent API in the pages read (10 of 20).\n\nThe checklist (https://www.anchorterminal.com/benchmark/#checklist-transparency):\n\n- 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.\n- 0 to 30, data handling and retention statements that agree with each other (privacy policy, DPA, retention periods, subprocessors).\n- 0 to 20, a deprecation policy or notices with dates.\n- 0 to 20, telemetry disclosed with an opt-out (local software), or subprocessors and data locations disclosed (hosted).\n\nThe other half of Transparency and trust is the provenance score, computed from checked facts (below). The category score is the mean of the two.\n\nProvenance checks not met in full (half of this category, computed from checked facts):\n\n- Domain age: assemblyai.com, registered 2016-12-24 (9 years) (11 of 15)\n- Terms of service: read, states 7 of the 7 things a reader expects, and has 2 clauses that cost points (6 of 10)\n- Privacy policy: read, states 7 of the 8 things a reader expects (9.3 of 10)\n- security.txt: could not be fetched (0 of 10)\n\n## 7. Schema \u0026 documentation, 83 out of 100, up to 2.8 more on the total\n\nWhy it scored 83: An AsyncAPI 3.0 file describes the socket and an OpenAPI 3.0.3 file the token endpoint. No description file was found for the agents, sessions and webhook REST endpoints, and the AsyncAPI file lacks `language_codes`, `voice_focus`, `transcript.agent.delta` and `is_error`, which the docs added on 23 July 2026 (19 of 25). llms.txt, a Markdown copy of each page and a docs MCP server (10). Operations and fields say what they are for and when to use them, such as when to choose `hold` over `interactive` execution (17 of 20). Enums, ranges and defaults on audio encoding, turn detection and timeouts. Tool `parameters` are a JSON Schema the server does not validate at `session.update` (12 of 15). Examples in cURL, Python and JavaScript, and tables of session error codes with close codes (13 of 15). A dated public changelog tagged by product and a `/v1` path, with no written versioning policy found (12 of 15).\n\nThe checklist (https://www.anchorterminal.com/benchmark/#checklist-schema):\n\nAPIs and MCP servers.\n\n- 25, a machine-readable contract (a public OpenAPI file or similar; for MCP, typed JSON Schema inputs on every tool).\n- 10, llms.txt or Markdown docs served for agents.\n- 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.\n- 0 to 15, typed inputs with enums, constraints and required fields, and no free-form JSON blobs.\n- 0 to 15, examples and documented error responses.\n- 15, versioning and a public changelog.\n\nModels 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.\n\n## What we couldn't check\n\nWhat 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.\n\n- The Terms of Service (effective 1 July 2026) bar competitive analysis and benchmarking (2.4(f)), any automated or programmatic method of extracting data or Output (4.4), and use of the Services or Output for automated decision-making (2.4(j)). Recorded as a fact with no deduction. It matters before any probe is run.\n- unchecked: the concurrent-session limit for Voice Agent accounts. No figure is in the Voice Agent docs, and the published streaming limits are not said to apply.\n- unchecked: security.txt. robots.txt on www.assemblyai.com disallows `/.well-known`, so `securityTxt` is unknown.\n- unchecked: the subprocessor list and the trust centre, which redirect to app.vanta.com and were not read.\n- unchecked: the Data Processing Addendum, the acceptable use policy and the biometric data addendum were not read.\n- unchecked: how long Voice Agent recordings and timelines are kept. The retention page has no Voice Agent section.\n- unchecked: the EU host for the Voice Agent API. The troubleshooting page mentions an EU endpoint without naming it.\n- unchecked: GitHub stars and package download counts, so `popularity` is empty.\n- Whether the realtime speech-to-text incidents of 31 July and 6 October 2026 affected Voice Agent sessions is not stated on the status page.\n- The RDAP host answered 400 to a robots.txt request, and one domain lookup was made after that.\n- The docs carry text addressed to AI models (a note to fetch the docs index, an integration prompt and an `AGENTS.md`). It was treated as data.\n\n## Weaknesses\n\n- No number is published for the concurrent-session limit that the `concurrency_exceeded` error enforces\n- status.assemblyai.com lists the Asynchronous API, Streaming API and LLM Gateway, with no Voice Agent component\n- Telephony is inbound only, over a SIP trunk on a Twilio account the customer owns\n- API keys have no scopes or expiry, and opting out of model training needs a paid plan\n- The Terms of Service bar benchmarking and competitive analysis (2.4(f)) and any automated or programmatic method of extracting data or Output (4.4), which matters before any probe is run\n\n## What costs an agent a turn today\n\nThe notes we give agents before they call it. Each one is a workaround an agent shouldn't need.\n\n- Send `session.end` before closing the socket. Without it the session stays resumable, and billed, for 30 seconds\n- Retry only `at_capacity`, `concurrency_exceeded` and `internal_error`, with backoff. Every other session error code is fatal\n- Send `tool.result` inside the `reply.done` handler, not straight after `tool.call`\n- Mint a one-time token with `GET /v1/token` for browser clients and never ship the API key to a page\n- Deduplicate webhooks on `event_id`, and verify `X-AAI-Signature` against the raw body bytes because a 401 is not retried\n\n## When it's done\n\nSend 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.\n",
    "grade": "B",
    "score": 64.4,
    "assessed": "2026-10-09",
    "run": "October 2026 research run",
    "categories": [
      {
        "key": "reliability",
        "name": "Reliability",
        "score": 60,
        "maxGain": 8,
        "reason": "Scored with the hosted lines. Statuspage at status.assemblyai.com with component history, but its components are the Asynchronous API, Streaming API and LLM Gateway and none is the Voice Agent API (15 of 20). No incident in the 90 days to 9 October 2026 names the Voice Agent API. Five name the realtime speech-to-text it runs on, including about two hours of degraded Universal-3.5 Pro streaming in us-east on 31 July and about two hours of rejected Universal-3.6 Pro Realtime sessions on the US endpoint on 6 October. Whether agent sessions were affected is not stated (15 of 30). A per-account concurrent-session limit is enforced and no figure is published for it (0 of 15). The events reference names `at_capacity`, `concurrency_exceeded` and `internal_error` as the only retryable codes and says to reconnect with backoff, sessions resume for 30 seconds, and webhooks carry a stable `event_id`. No idempotency key on `POST /v1/agents` (10 of 15). A Service Level Agreement of 99.9 per cent a month with credits is published, written against an Order Form (10). The launch note of 29 April 2026 says the API is available to all customers and no beta label was found (10).",
        "checklist": [
          "Hosted APIs, MCP servers, models and platforms.",
          "- 20, a public status page with component history (Statuspage, Instatus, BetterStack or the vendor's own).\n- 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.\n- 15, rate limits documented with numbers.\n- 15, documented 429 or overload handling (Retry-After, backoff guidance), and idempotency keys or safe-retry guidance where writes are involved.\n- 10, an SLA published for any paid tier.\n- 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.\n- 25, a public CI and test suite, passing on the default branch.\n- 0 to 25, open crash or regression issues relative to activity (25 for few and handled, 0 for many, old and unanswered).\n- 15, semver discipline and breaking changes called out in a changelog.\n- 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."
        ],
        "checklistUrl": "https://www.anchorterminal.com/benchmark/#checklist-reliability"
      },
      {
        "key": "payments",
        "name": "Payments \u0026 pricing",
        "score": 40,
        "maxGain": 7.5,
        "reason": "No x402, MPP or L402 (0 of 40). One per-unit price on the public pricing page, $4.50 an hour billed per second (20). $50 of credit with no card, which the billing docs say covers the Voice Agent API (20). A person has to sign up in a browser to get the first key, and no key-creation API was found (0).",
        "checklist": [
          "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.\n- 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.\n- 20, a free tier or trial that doesn't need a card.\n- 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."
        ],
        "checklistUrl": "https://www.anchorterminal.com/benchmark/#checklist-payments"
      },
      {
        "key": "security",
        "name": "Security \u0026 auth",
        "score": 62,
        "maxGain": 6.7,
        "reason": "Plain API keys per project, created and deleted in the dashboard, with no scopes or expiry found (20 of 30). Browser clients pass a token in the `token` query parameter. We took no deduction because the token is one-time and redeemable for at most 600 seconds, which is a judgement call. A read-only Reader role exists for dashboard members but not for keys. HTTP tool header values are write-only and encrypted at rest, tool calls go only to public `https` hosts with redirects refused, and the docs show how to gate tools by conversation state. No built-in confirmation step before a tool that writes (10 of 20). Caller speech and tool results reach a model. Tool arguments are drawn only from user turns and tool results since 26 August 2026 and tool responses are capped, but no prompt-injection guidance was found (8 of 15). Sessions keep a timeline with tool calls, webhook deliveries are logged per session, usage is broken down by key, and the dashboard has activity logs (12 of 15). The security page states a SOC 2 Type 2 report and third-party penetration tests. No bug bounty or disclosure policy was found, and robots.txt closes `/.well-known`, so security.txt went unread (12 of 20).",
        "checklist": [
          "- 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.\n- 0 to 20, read-only or least-privilege modes, and confirmation or approval for destructive actions.\n- 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.\n- 0 to 15, audit logs or per-call visibility for the operator.\n- 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."
        ],
        "checklistUrl": "https://www.anchorterminal.com/benchmark/#checklist-security"
      },
      {
        "key": "ergonomics",
        "name": "Agent ergonomics",
        "score": 72,
        "maxGain": 4.6,
        "reason": "Scored as an API. Events are small typed JSON, HTTP tool responses are capped at 8 KiB before they reach the model, and the session list leaves artefacts out (18 of 25). `GET /v1/sessions` pages by `cursor` with `limit` from 1 to 200 and filters on `status` and `agent_id` (18 of 20). Errors carry `code`, `message` and sometimes `param`, and REST validation errors name the field (17 of 20). No idempotency keys on REST writes. Sessions resume within 30 seconds, webhooks dedupe on `event_id`, and `hold` mode silences the agent during long tools (10 of 20). An agent needs three fields (`name`, `system_prompt`, `voice`). The official Python and Node SDKs have no Voice Agent client, so callers write JSON over a socket or start from the two starter repositories (9 of 15).",
        "checklist": [
          "- 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).\n- 20, pagination, filtering and output-size controls.\n- 20, actionable, documented error responses, codes and messages an agent can recover from.\n- 20, idempotency or safe retries, and for MCP the `readOnlyHint` and `destructiveHint` annotations.\n- 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."
        ],
        "checklistUrl": "https://www.anchorterminal.com/benchmark/#checklist-ergonomics"
      },
      {
        "key": "maintenance",
        "name": "Maintenance \u0026 community",
        "score": 64,
        "maxGain": 3.2,
        "reason": "Universal-3.6 Pro Realtime, the speech-to-text the pricing page lists inside the Voice Agent API, shipped on 29 September 2026. The last changelog entry tagged Voice Agent API is 26 August 2026. We scored on the August entry, the last one tagged for this API, 44 days before the check (20). Ten entries tagged Voice Agent API between 13 July and 26 August 2026 (20). A public changelog that records fixes, a support address and a Discord link we did not test (11 of 15 for a closed service). The Python SDK (1.6.1) and Node SDK (4.41.5) were tagged on 24 September 2026 but have no Voice Agent client. The two starter repositories were last changed on 16 August 2026 (8 of 15). The starters need no dependencies and have no licence file, tests or CI (5 of 10).",
        "checklist": [
          "- 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.\n- 20, at least three releases or dated changelog entries in the last 90 days.\n- 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.\n- 15, presence in the official MCP registry under a verified namespace (MCP servers), or current official SDKs (APIs and models).\n- 10, package health, current dependencies and CI.",
          "Models are read for deprecation notice periods and model churn rather than release counts."
        ],
        "checklistUrl": "https://www.anchorterminal.com/benchmark/#checklist-maintenance"
      },
      {
        "key": "transparency",
        "name": "Transparency \u0026 trust",
        "score": 66,
        "maxGain": 3,
        "reason": "A closed service with a clear service agreement. The starter repositories have no licence file (15 of 30). The Terms of Service grant a licence to train on Customer Data by default, the opt-out needs a paid plan, and the retention page gives periods for streaming, sync, dictation, asynchronous and LLM Gateway data but none for Voice Agent sessions, which are recorded and stored. A DPA is published and the privacy policy states Data Privacy Framework certification (14 of 30). The changelog dates deprecations ahead of time, such as the 11 July 2026 notice of a change on 7 August, and the terms promise reasonable prior notice of major changes. No written deprecation policy was found (12 of 20). The subprocessor list sits on a third-party trust centre we did not read. US and EU endpoints are mentioned, with no EU host named for the Voice Agent API in the pages read (10 of 20).",
        "blend": "editorial 51, provenance 81",
        "checklist": [
          "- 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.\n- 0 to 30, data handling and retention statements that agree with each other (privacy policy, DPA, retention periods, subprocessors).\n- 0 to 20, a deprecation policy or notices with dates.\n- 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."
        ],
        "checklistUrl": "https://www.anchorterminal.com/benchmark/#checklist-transparency"
      },
      {
        "key": "schema",
        "name": "Schema \u0026 documentation",
        "score": 83,
        "maxGain": 2.8,
        "reason": "An AsyncAPI 3.0 file describes the socket and an OpenAPI 3.0.3 file the token endpoint. No description file was found for the agents, sessions and webhook REST endpoints, and the AsyncAPI file lacks `language_codes`, `voice_focus`, `transcript.agent.delta` and `is_error`, which the docs added on 23 July 2026 (19 of 25). llms.txt, a Markdown copy of each page and a docs MCP server (10). Operations and fields say what they are for and when to use them, such as when to choose `hold` over `interactive` execution (17 of 20). Enums, ranges and defaults on audio encoding, turn detection and timeouts. Tool `parameters` are a JSON Schema the server does not validate at `session.update` (12 of 15). Examples in cURL, Python and JavaScript, and tables of session error codes with close codes (13 of 15). A dated public changelog tagged by product and a `/v1` path, with no written versioning policy found (12 of 15).",
        "checklist": [
          "APIs and MCP servers.",
          "- 25, a machine-readable contract (a public OpenAPI file or similar; for MCP, typed JSON Schema inputs on every tool).\n- 10, llms.txt or Markdown docs served for agents.\n- 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.\n- 0 to 15, typed inputs with enums, constraints and required fields, and no free-form JSON blobs.\n- 0 to 15, examples and documented error responses.\n- 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."
        ],
        "checklistUrl": "https://www.anchorterminal.com/benchmark/#checklist-schema"
      }
    ],
    "provenance": [
      {
        "label": "Domain age",
        "value": "assemblyai.com, registered 2016-12-24 (9 years)",
        "points": 11,
        "max": 15
      },
      {
        "label": "Terms of service",
        "value": "read, states 7 of the 7 things a reader expects, and has 2 clauses that cost points",
        "points": 6,
        "max": 10
      },
      {
        "label": "Privacy policy",
        "value": "read, states 7 of the 8 things a reader expects",
        "points": 9.3,
        "max": 10
      },
      {
        "label": "security.txt",
        "value": "could not be fetched",
        "points": 0,
        "max": 10
      }
    ],
    "unchecked": [
      "The Terms of Service (effective 1 July 2026) bar competitive analysis and benchmarking (2.4(f)), any automated or programmatic method of extracting data or Output (4.4), and use of the Services or Output for automated decision-making (2.4(j)). Recorded as a fact with no deduction. It matters before any probe is run.",
      "unchecked: the concurrent-session limit for Voice Agent accounts. No figure is in the Voice Agent docs, and the published streaming limits are not said to apply.",
      "unchecked: security.txt. robots.txt on www.assemblyai.com disallows `/.well-known`, so `securityTxt` is unknown.",
      "unchecked: the subprocessor list and the trust centre, which redirect to app.vanta.com and were not read.",
      "unchecked: the Data Processing Addendum, the acceptable use policy and the biometric data addendum were not read.",
      "unchecked: how long Voice Agent recordings and timelines are kept. The retention page has no Voice Agent section.",
      "unchecked: the EU host for the Voice Agent API. The troubleshooting page mentions an EU endpoint without naming it.",
      "unchecked: GitHub stars and package download counts, so `popularity` is empty.",
      "Whether the realtime speech-to-text incidents of 31 July and 6 October 2026 affected Voice Agent sessions is not stated on the status page.",
      "The RDAP host answered 400 to a robots.txt request, and one domain lookup was made after that.",
      "The docs carry text addressed to AI models (a note to fetch the docs index, an integration prompt and an `AGENTS.md`). It was treated as data."
    ],
    "weaknesses": [
      "No number is published for the concurrent-session limit that the `concurrency_exceeded` error enforces",
      "status.assemblyai.com lists the Asynchronous API, Streaming API and LLM Gateway, with no Voice Agent component",
      "Telephony is inbound only, over a SIP trunk on a Twilio account the customer owns",
      "API keys have no scopes or expiry, and opting out of model training needs a paid plan",
      "The Terms of Service bar benchmarking and competitive analysis (2.4(f)) and any automated or programmatic method of extracting data or Output (4.4), which matters before any probe is run"
    ],
    "agentNotes": [
      "Send `session.end` before closing the socket. Without it the session stays resumable, and billed, for 30 seconds",
      "Retry only `at_capacity`, `concurrency_exceeded` and `internal_error`, with backoff. Every other session error code is fatal",
      "Send `tool.result` inside the `reply.done` handler, not straight after `tool.call`",
      "Mint a one-time token with `GET /v1/token` for browser clients and never ship the API key to a page",
      "Deduplicate webhooks on `event_id`, and verify `X-AAI-Signature` against the raw body bytes because a 401 is not retried"
    ],
    "recheck": "https://www.anchorterminal.com/builders/#disputes"
  },
  "meta": {
    "attribution": "Anchor Terminal (https://www.anchorterminal.com)",
    "docs": "https://www.anchorterminal.com/docs/",
    "generatedAt": "2026-10-10",
    "license": "CC-BY-4.0",
    "method": "https://www.anchorterminal.com/benchmark/",
    "methodology": "0.4",
    "openapi": "https://www.anchorterminal.com/openapi.json",
    "preview": false,
    "run": "2026-10-01",
    "runLabel": "October 2026 research run"
  }
}