Configuration Reference

Enable only the capabilities you need. This reference follows the released ChatVoiceConfig schema. Keep credentials in server-side ChatEnv; the browser receives sanitized readiness only.

Capability	Primary settings	Guide
Transcription	ASR channel and local runtime or HTTP endpoint	Web guide
Notes, revision, Todo	Notes base, key and model	Text models
Automatic titles	Title base, key and model	Text models
System speech	TTS protocol, endpoint, key, model and voices	TTS
Voice cloning	VoiceClone URL	Voice cloning
Realtime	Legacy audio credentials and valid model entitlement	Web guide
Copilot preview	Feature flag, notes text model and ASR	Web guide

Use ChatEnv

Install ChatVoice before provider discovery. Its canonical storage namespace is ChatVoice and its command alias is chatvoice.

chatenv status
chatenv init -t chatvoice -I
chatenv paste --stdin --profile speakr --yes

The last command reads KEY=VALUE text from standard input. Paste your values, finish input, then activate the named profile:

chatenv use speakr -t chatvoice -I
chatenv test -t chatvoice -I

The probe makes real text/TTS requests and can consume quota. Reload the service to apply startup configuration changes. ChatEnv does not automatically export variables to unrelated processes; runtime path overrides, CLI passwords and read tokens must also be supplied by the caller where required.

Independent text models

Field	Meaning
CHATVOICE_MEETING_NOTES_API_BASE	OpenAI-compatible base; /chat/completions is appended
CHATVOICE_MEETING_NOTES_API_KEY	Sensitive notes credential
CHATVOICE_MEETING_NOTES_MODEL	Notes model, also used for revisions and Todo
CHATVOICE_MEETING_TITLE_API_BASE	Independent title base
CHATVOICE_MEETING_TITLE_API_KEY	Sensitive title credential
CHATVOICE_MEETING_TITLE_MODEL	Title model identifier

Setting an independent base or key requires the complete base/key/model triple for that purpose. You may explicitly configure equal values for both purposes, but missing values are never borrowed. Todo needs no additional model setting. The in-meeting Copilot fast answer path also reuses the notes model; missing notes configuration returns HTTP 503 and never borrows voice, title or browser-supplied arbitrary model settings.

CHATVOICE_MEETING_NOTES_API_BASE=https://model.example.com/v1
CHATVOICE_MEETING_NOTES_API_KEY=[REDACTED]
CHATVOICE_MEETING_NOTES_MODEL=notes-model
CHATVOICE_MEETING_TITLE_API_BASE=https://model.example.com/v1
CHATVOICE_MEETING_TITLE_API_KEY=[REDACTED]
CHATVOICE_MEETING_TITLE_MODEL=title-model

All example addresses, keys and model names are placeholders.

Copilot preview

Field	Default	Meaning
CHATVOICE_COPILOT_ENABLED	0	Enables /copilot, app navigation and /api/copilot/*
CHATVOICE_COPILOT_AUTO_PREPARE	0	Allows speculative draft preparation; off by default, manual Submit remains available
CHATVOICE_COPILOT_THINKING_MODE	provider-default	Fast-answer thinking policy: provider-default adds no vendor field; ark-disabled sends Volcengine Ark thinking.type=disabled only for the Copilot request; other values fail closed
CHATVOICE_ENV_PROFILE	Empty	Load a named ChatEnv ChatVoice profile without changing the global active profile; missing names do not fall back to active

Uploads accept TXT, Markdown, basic text PDFs and DOCX; URL fetch is not supported. Files are capped at 2 MiB, with declared and chunked body limits before multipart parsing. Answer/prepare JSON bodies are limited to 1 MiB before parsing. Authentication, CSRF and the preview flag are checked before the material form is parsed.

PDF parsing uses pypdf>=6.19,<7.0 and validates direct-length streams, escaped names and decode limits before extraction. No-filter, Flate, ASCII85, ASCIIHex and RunLength streams are supported within those limits. Indirect lengths, nested streams, compressed object/xref streams, Form references, custom decode parameters and unsupported image/encryption/LZW filters fail closed. Re-export complex files as basic text PDFs or use TXT/Markdown. Scanned PDFs with no text do not receive OCR.

Copilot explicitly requests max_tokens=384 and checks SSE line/event/aggregate bytes, event count, output length, a 60-second elapsed deadline and a 15-second socket-read timeout. The token parameter is not a spending guarantee. Disconnect, logout and session invalidation stop processing at cooperative read boundaries, close the response and never produce done. Partial output is not completion, and cancellation cannot reverse quota already spent. Owner and cookie/CSRF boundaries remain unchanged.

ASR

Field	Default	Meaning
CHATVOICE_ASR_CHANNEL	api-server when an endpoint exists; otherwise stub-local	api-server, funasr-gpu, funasr-cpu or stub-local
CHATVOICE_ASR_API_URL	Empty	Complete transcription endpoint
CHATVOICE_ASR_API_KEY	Empty	Optional sensitive Bearer credential
CHATVOICE_ASR_PREWARM	1	Prewarm the selected persistent local FunASR channel
CHATVOICE_FUNASR_ALLOW_SUBPROCESS_WORKER	0	Debug compatibility only; short-lived workers reload models per request

Local FunASR also reads process variables FUNASR_MODEL (default iic/SenseVoiceSmall) and FUNASR_GPU_DEVICE (default cuda:0). They are not fields in the current typed ChatEnv schema. CPU uses the CPU; GPU requires compatible PyTorch/CUDA/FunASR dependencies. The stub channel is not real recognition.

System TTS

Field	Meaning
CHATVOICE_TTS_API_TYPE	openai, volcengine or qwen
CHATVOICE_TTS_API_BASE	HTTP base; exact ws/wss endpoint for Qwen
CHATVOICE_TTS_API_KEY	Sensitive independent TTS credential
CHATVOICE_TTS_MODEL	Model identifier
CHATVOICE_TTS_RESOURCE_ID	Volcengine resource ID, separate from model
CHATVOICE_TTS_VOICES	JSON list of id/label; first voice is default

Any nonempty independent field enables completeness checks. The resource ID is not required for openai or qwen; other required fields cannot be omitted. The Qwen path retains its sk-sp guard, which is not applied to other protocols. See TTS examples and endpoint paths.

Voice cloning and runtime

Field	Default	Meaning
CHATVOICE_VOICECLONE_URL	Empty	VoiceClone service base
CHATVOICE_VOICECLONE_TIMEOUT_SECONDS	180	Request timeout in seconds
CHATVOICE_HOME	$CHATARCH_HOME/chatvoice or ~/.chatarch/chatvoice	Runtime root; provide a process override consistently for CLI and service launcher
CHATVOICE_SQLITE_PATH	data/meetings.sqlite3 under runtime root	Explicit database path supplied through the process environment
CHATVOICE_REALTIME_MODELS	Empty	Comma-separated allowlist, subject to backend support and entitlement

CHATARCH_HOME is the shared ChatEnv home setting. See runtime layout for path precedence and backup.

Legacy deployment compatibility

These fields remain supported. Prefer independent text/TTS configuration for new deployments; Todo does not require legacy selectors.

Field	Compatibility purpose
CHATVOICE_OPENAI_API_BASE	Legacy audio account HTTP base; defaults to the Token Plan compatible-mode endpoint
CHATVOICE_OPENAI_API_KEY	Sensitive legacy system-TTS/realtime credential; requires sk-sp
CHATVOICE_OPENAI_API_MODEL	Legacy text default when independent settings are absent; qwen3.7-plus
CHATVOICE_MEETING_NOTES_PROVIDER	Legacy notes selector; defaults to token-plan-chat-completions
CHATVOICE_MEETING_NOTES_CRS_PROFILE	Legacy external configuration reference
CHATVOICE_MEETING_NOTES_CRS_API_BASE	Legacy explicit base override
CHATVOICE_MEETING_NOTES_CRS_API_KEY	Sensitive legacy key override

Independent text configuration takes precedence and fails if partial; it does not borrow these fields. Plan quota, API protocol and entitlement are separate concerns. Do not switch to usage-billed endpoints as an implicit exhaustion fallback.