Anam AI Avatars

Creates a link that opens a persona in a browser and returns it with its `url`. Anyone with the link can talk to the persona until it expires (`expiresInHours`, 168 by default) or runs out of uses (`usageLimit`, 10 by default).

anam_share_link_create

Create a persona the user can talk to now, in one call, from your prompt or a common role: Anam’s recommended avatar, voice and LLM unless you pass IDs, `systemPrompt` or else the `template`’s prompt, and a share link (up to 10 sessions over 7 days, on the organization’s minutes). Returns the persona, the link’s URL and its Lab page.

anam_quickstart

Create a one-shot avatar from a public image URL (`imageUrl`); this server cannot read files from the user’s computer. Image moderation and creation capacity limits apply. Only use images of people who have consented. Check `status` with anam_avatar_get after about 10 seconds.

anam_avatar_create

Deletes a custom LLM. While personas (or their drafts) use it, the delete is refused with 409, since their sessions would fail: move them to another LLM with anam_persona_update first.

anam_llm_delete

Deletes a knowledge document and its search index, so personas searching its folder no longer find its content.

anam_knowledge_document_delete

Deletes a knowledge folder and every document in it. Detach it from personas first with anam_persona_knowledge_remove.

anam_knowledge_folder_delete

Deletes a persona; one that is missing or already deleted returns 404. Its tools aren’t deleted. A `Knowledge_<persona>` tool from anam_persona_knowledge_add is left unused: delete it with anam_tool_delete unless another persona uses it.

anam_persona_delete

Deletes a persona tool. Its `usageCount` (anam_tool_get) says how many personas use it; take it out of their `toolIds` first. Anam’s system tools can’t be deleted.

anam_tool_delete

Archives a pronunciation dictionary for good: it leaves the API and Lab, and sessions can’t select it. Its entries and history are kept until the organization is deleted. One a persona uses is refused with 409 `DICTIONARY_ATTACHED` and the persona IDs; remove it from those personas in Lab first.

anam_pronunciation_dictionary_delete

Permanently delete customer data from an ended session. `scope` `all` (default): the report (transcript, analytics, tool calls), insights, summary, recording, initial message and requester IP; `recording`: the video only; `report`: the report, insights and summary only. The session record and usage stay for billing. Irreversible; repeating it succeeds.

anam_session_delete

Deletes a share link; its `url` stops working.

anam_share_link_delete

Deletes a voice from Anam, then from its provider. While personas (or their drafts) use it, the delete is refused with 409, since their sessions would have no voice: move them to another voice with anam_persona_update first.

anam_voice_delete

Deletes an avatar. By default the delete is soft: the avatar is hidden, with no restore in Lab or the API, and its files are kept. `hard: true` also erases its stored image and versions (for data-erasure requests; it also purges a soft-deleted one), except assets another avatar uses. A hard delete is refused with 409 and the persona IDs while personas use the avatar: move them with anam_persona_update first.

anam_avatar_delete

Cancels the job if it is still running, then deletes its recording, session report and job content; the job stays, without content, and can’t be retried. Returns 200 once everything is deleted, or 202 while Lab finishes in the background, retrying every minute until it succeeds. Repeating the call is safe.

anam_avatar_video_delete

Find calls by date, persona and what was said. Without `text`, returns a compact list of sessions, newest first. With `text`, also reads the transcripts of up to 25 of the newest matching sessions and returns those that mention it, with snippets; narrow the date range to search further back. With `sessionId`, reads that session’s whole transcript instead, 50 messages at a time (`limit`, up to 100): pass the returned `nextOffset` as `offset` until it is null.

anam_session_search

Creates an asynchronous avatar-video render backed by a normal recorded Anam session. A saved persona contributes only its snapshotted avatar, voice, language, model and director notes; its greeting, brain, prompt and tools never run. The request is rejected rather than queued when either concurrency limit is full. Send exactly one of these shapes: 1. `personaId`; 2. `avatarId`, `voiceId`.

anam_avatar_video_create

Returns a knowledge document: file name, type, size, folder (`knowledgeFolderId`), chunking and `status`. Only `READY` documents are searchable; a `FAILED` one has `errorMessage`.

anam_knowledge_document_get

Returns `downloadUrl`, a link to the document’s original file that works for an hour, and its `filename`.

anam_knowledge_document_download

Returns a knowledge folder: name, description, document count and chunking settings.

anam_knowledge_folder_get

Returns a meeting invite: meeting URL and provider, `status` (`pending`, `active`, `ended`, `cancelled` or `failed`, with `statusReason`), `joinState`, join time and, once the persona joins, `sessionId`.

anam_meeting_invite_get

Returns a persona’s full configuration: system prompt (`brain`), avatar, voice, LLM, `toolIds` and tools, knowledge, share links and session settings.

anam_persona_get

Returns a persona tool: name, description, type, config and `usageCount`. A webhook config shows its structure, not its values: header values, URL IDs and tokens, query values and body template values show as •••• (a `Bearer` scheme, `{{placeholders}}`, header names, JSON keys and the host stay). Send them back unchanged to keep them.

anam_tool_get

Returns saved widget display configuration and avatar media. Origin access controls are omitted. Missing or temporarily unavailable idle previews are null; clients should show avatarImageUrl instead. Preview availability does not indicate whether the avatar can start a call.

anam_persona_widget_get

Returns a pronunciation dictionary with its entries and current revision.

anam_pronunciation_dictionary_get

Returns a session and, once it has ended and its report has arrived, its `toolCalls` (never for LiveKit or audio-passthrough sessions): names, timing, status and errors, with arguments and results only when transcripts were on and the session isn’t zero-data-retention. What was said is in anam_session_transcript_get.

anam_session_get

Returns a presigned URL to download the session’s recording. Zero-data-retention sessions have no recording to return.

anam_session_recording_get

Returns a session’s conversation, 50 messages at a time unless `limit` (up to 100) says otherwise: pass the returned `nextOffset` as `offset` until it is null. `totalMessages` counts them all. Zero-data-retention sessions have no transcript.

anam_session_transcript_get

Returns per-turn latency for a session, where time went from transcription through the LLM and TTS to first audio, and session totals. It is 404 until the session has ended and its report has arrived, and always for LiveKit and audio-passthrough sessions. Message text needs `includeMessages: true` and transcripts on.

anam_session_analytics_get

Returns a share link: its public `url`, persona, expiry, usage, allowed origins and whether it is enabled.

anam_share_link_get

Returns a voice: provider and model, gender, accent (`country`), description and sample URL. Its `id` is a persona’s `voiceId`.

anam_voice_get

Returns an LLM: format, model, temperature, token limit, reasoning settings and, for a custom one, `urls`. API keys are never returned.

anam_llm_get

Returns an avatar: `status` (`processing` until its preview videos exist, then `ready`), name, description, tags, style, image and versions. Its `id` is a persona’s `avatarId`.

anam_avatar_get

Returns an avatar video job: its `status` and, once `completed`, `content.url`, a link to the video that works for an hour (`content.expiresAt`); read the job again for a new one. A `failed` job’s `failure` says why and whether a retry can fix it (`retryable`).

anam_avatar_video_get

Returns the organization’s concurrent-session limit, about how many sessions are in use, whether a new one can likely start and an estimated wait for a free slot. It is advisory: the count can briefly over-report, and a start can still be refused with 429.

anam_session_concurrency_get

Returns latency percentiles (p50, p90, p95, p99, max, avg) for each stage from transcription through the LLM and TTS to first audio, interruption and error rates, tool call outcomes and the slowest turns, across sessions. Only sessions where Anam runs the conversation (turnkey, custom LLM, ElevenLabs agent) have analytics; LiveKit and audio-passthrough ones count in `sessionCount` only. It reads the 1,000 newest matching sessions with analytics; `truncated` says more matched. The last 24 hours by default, up to 31 days. anam_session_insights summarises the same with findings.

anam_session_stats

Starter files that put a persona in the user’s app, to write yourself: a server route that swaps the `ANAM_API_KEY` environment variable for a short-lived session token, a page using `@anam-ai/js-sdk`, an env file without the key, and install and run commands. `express` and `fastapi` come with a Vite page.

anam_scaffold

Let a persona search more knowledge folders during calls. A persona searches its knowledge folders through a knowledge tool, as in Lab’s Build page; if other personas use the same tool, they change too. The persona’s other tools are kept.

anam_persona_knowledge_add

Concise, current steps for putting a persona into the user’s app: the server-side session token pattern, the package to install, a minimal snippet and the guides to follow, plus the start of the stack’s main docs page. A persona ID fills it in.

anam_integration_guide

Creates a pronunciation dictionary from a Cartesia pronunciation dictionary file: an array of `{ text, pronunciation, case_sensitive? }` rows, with an optional Anam `sounds_like` fallback. A pronunciation containing `<<` or `>>` is imported as an IPA entry, anything else as a sounds-like one. The file must hold 1 to 500 rows in at most 1 MiB of UTF-8 JSON (an array is measured without whitespace). If any row is invalid, nothing is saved and every error is listed by row.

anam_pronunciation_dictionary_import

Imports a provider voice and returns it; its `id` is a persona’s `voiceId`. The voice is saved once a speech sample generates with it: up to 95 seconds, and about a minute for a first ElevenLabs Voice Library import. `displayName` and `description` default to the provider’s. Only use voices of people who have consented. An import creates a new voice and generates a speech sample on Anam’s provider account, or with `useOrganizationKey: true` on the organization’s saved ElevenLabs key and its credits (for its own clones and designed voices). Retrying creates another voice, so check `anam_voice_list` first. A `shareUrl` (instead of `provider` and `providerVoiceId`) imports a Professional Voice Clone shared by link; one the organization already imported on that model gets a new sample instead of a new voice.

anam_voice_import