Rime Prism

# Fix list: Rime Prism

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/rime-prism, the October 2026 research run, assessed 28 October 2026. Grade C, 57.9 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 Rime Prism: 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. Reliability, 41 out of 100, up to 11.8 more on the total

Why it scored 41: Graded as a hosted API on its launch day. status.rime.ai redirects to an incident.io page at rimestat.us with Text-to-speech API and Platform API components, but no Prism component on 28 October 2026 (10 of 20). Prism launched that day, so there is no incident history for it. The two incidents on the page, on 3 and 5 September, were text-to-speech partial degradations (5 of 30). The errors page publishes per-session numbers (55 seconds a turn, 60 seconds of queued speech, 120 seconds or 512 tokens a reply, 5 minutes idle, 1 MiB a message), but no concurrency limit per account or key, and a full server answers 503 (8 of 15). Overload is documented as `capacity_exceeded`, `unavailable`, a 503 refusal and close code 1001, each with a retry-after-backoff instruction, and the tools guide says to make each tool safe to run once and never rerun a call with an unknown outcome. No Retry-After header is mentioned (13 of 15). SLAs are an Enterprise item and none is published (0). The changelog calls Prism available, but all four SDKs Rime recommends are 0.1.0 alpha releases, the Pipecat path uses a pinned preview, and the pricing page says half duplex at launch (5 of 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.

## 2. Security & auth, 56 out of 100, up to 7.7 more on the total

Why it scored 56: Graded as a hosted API. One Rime API key per request, sent as a Bearer header on the upgrade. Prism reads no query parameters and the docs say not to put the key in the URL. No scopes, no read-only keys and no ephemeral tokens for client devices were found, the same key also covers text-to-speech and speech-to-text, and rotation isn't described beyond creating tokens in the dashboard (20 of 30). `tool_choice` is `auto` only, so a backend can't force a call, but it can remove a tool mid-call with `session.update`. The prompting guide says Prism can call a tool before the caller confirms and to keep confirmation rules in tool code. The launch pages say every reply is text before speech so guardrails can check it, but the docs describe no step that holds a reply for checks (10 of 20). Prism takes untrusted caller audio and tool results. The docs say to confirm outcomes from the agent's own records, not the reply text, with no explicit prompt-injection guidance (6 of 15). Request, session and correlation IDs for support and token usage on each `response.done`. No per-call log or audit trail in the dashboard was found (6 of 15). SOC 2 Type II since May 2025, HIPAA compliance since February 2024 with an audit in March 2026, and a vulnerability disclosure policy linked from the security page. security.txt answered 404 and no bug bounty was found (14 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. Payments & pricing, 40 out of 100, up to 7.5 more on the total

Why it scored 40: No x402, MPP or L402 (0). $0.08 a minute on Flex and $0.064 on Scale, published without a login (20). $50 of usage a month on the Free plan, Prism included, with no card (20). The free month for early sign-ups in Rime's launch announcement wasn't scored. A person signs up in a browser to get a key (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.

## 4. Schema & documentation, 69 out of 100, up to 5 more on the total

Why it scored 69: No OpenAPI or AsyncAPI file for the WebSocket API was found in llms.txt or the reference (0 of 25). llms.txt and a Markdown twin of every page (10). The overview says what Prism handles, what it doesn't (text-only replies, sampling settings, a choice of language model, ending the caller's turn) and when a pipeline fits better, and a chart says when to use the SDK, LiveKit, Pipecat or the raw WebSocket (18 of 20). The WebSocket reference types every client and server event field with defaults and allowed values, refuses unknown fields with `invalid_client_event` naming the field, and takes tool parameters as JSON Schema. The schemas exist only as tables, not a file (13 of 15). Python and Node.js examples on each guide, and an errors page that lists 41 codes with cause, scope and what to do, plus HTTP refusals and close codes (15). A dated changelog with an RSS feed, a `version` field in `session.created` (such as `20261023a`) and pinned SDK versions. The path carries `/v1`, with no version policy for the event set found (13 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.

## 5. Agent ergonomics, 73 out of 100, up to 4.4 more on the total

Why it scored 73: Scored as a model API. A 65,536-token context, with `truncation` set to leave out the oldest turns by default, a configurable `retention_ratio`, a `prism_truncation` report on each reply, `prism.conversation.get` to read the stored history and `conversation.item.delete` to trim it (20 of 25). Replies are capped at 120 seconds or 512 tokens and can't be set lower. `max_response_output_tokens`, `temperature` and text-only replies are refused. Up to 16 labels per reply for routing. No lists to page (8 of 20). Errors carry `code`, `scope`, `owner`, `param` and a `correlation_id`, and the docs tell clients to branch on code and scope, never on message (20). Client events are matched by `event_id`, a repeated identical tool result is acknowledged, and Prism records late results. Prism doesn't deduplicate requests, a dropped connection doesn't resume, and the client replays history into a new session (12 of 20). Short instructions are enough to start, the voice defaults to `astra` and interruption is on. Official SDKs in Python, Node.js, Go and Rust, all alpha, plus LiveKit and Pipecat quickstarts (13 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. Maintenance & community, 65 out of 100, up to 3.1 more on the total

Why it scored 65: Read as a closed model API. Prism's first release, version `20261023a`, went live on 28 October 2026 (30). One dated changelog entry for Prism, its launch, though the SDK repository published Prism-capable releases on 27 and 28 October (10 of 20). A public changelog with RSS, docs and community support on Free and email on paid plans, and a support address. Its responsiveness couldn't be judged on launch day (8 of 15). Current official SDKs in four languages, released on 28 October, all alpha with a warning that a release can change the API (12 of 15). The SDK repository is MIT, created 17 September 2026 and pushed on 28 October. Its CI and open issues weren't read (5 of 10). Rime retired Arcana from its cloud API 17 days after announcing it (29 July to 15 August 2026), the record Prism inherits as a Rime model.

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.

## 7. Transparency & trust, 70 out of 100, up to 2.6 more on the total

Made of editorial 58, provenance 81.

Why it scored 70: Closed service with clear terms from Rime Labs, Inc., which say Rime doesn't train on input or output except for a voice model the customer asks for, and which put beta services on an as-is basis. The SDK is MIT (15 of 30). The privacy policy (effective 22 August 2025) states zero retention by default, 90 days of server logs and no training on customer data, and the security page and docs agree. All three describe text in and audio out metered by characters, and none says how Prism's caller audio, transcripts or the stored conversation are handled, or how call data sent for fine-tuning is kept. The DPA at rime.ai/dpa couldn't be reached from our network (15 of 30). No written deprecation policy. Arcana's cloud retirement was dated in the changelog with 17 days' notice (10 of 20). A sub-processor list with locations, all in the United States, and the Prism endpoint stated as US East (us-east-2) (18 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: rime.ai, registered 2020-03-12 (6 years) (11 of 15)
- Terms of service: read, states 5 of the 7 things a reader expects, and has 1 clause that costs points (6.3 of 10)
- Privacy policy: read, states 3 of the 8 things a reader expects (6.3 of 10)
- security.txt: not found (0 of 10)

## 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.

- Response time and accuracy. Rime's docs and launch post give 0.57 seconds to first audio and a τ-Voice score of 56.6 on Artificial Analysis's harness, run by Rime, and the launch announcement the requester pasted says 300 ms and 4x faster than GPT-Live. We measured neither and couldn't find a public page with the 300 ms figure.
- How Prism's caller audio, transcripts and stored conversation are retained. The privacy policy, security page and docs describe text-to-speech only.
- The concurrency limit per account or key on each plan.
- Whether VPC and on-prem Prism exist today. The pricing page lists them for Enterprise, while the docs describe only the cloud endpoint.
- Whether 8 kHz phone audio is handled natively. The pricing page says telephony-ready at 8 kHz, while the docs say Prism takes 16 kHz PCM16 and the SDK converts other rates.
- When a Prism component will appear on the status page, and its incident record once it does.
- The DPA at rime.ai/dpa, which couldn't be reached from our network.

## Weaknesses

- Rime's status page had no Prism component on launch day, and there is no incident history for it
- No concurrency limit per account or key was found, and a full server refuses with 503
- All four SDKs are 0.1.0 alpha releases that may change their API
- English only, four voices, 16 kHz PCM16 in, with no mu-law, so phone calls need a bridge
- Privacy and security pages describe text in, audio out, and say nothing specific about caller audio

## 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.

- Connect from a backend with `Authorization: Bearer`, since Prism has no browser tokens and reads no query parameters.
- Send `session.update` before any audio, or every append fails with `session_not_initialized`.
- Keep streaming audio, silence included, while Prism speaks, because it finds turn ends and interruptions from that audio.
- Ask for the continuation after each tool result, since Prism doesn't resume on its own, and never rerun a call with an unknown outcome.
- Pin the exact SDK version, because the SDKs are alpha and a new release can change the API.

## 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.