FAQ — AI features

Last updated: May 19, 2026

FAQ — AI features

Common questions about Wevion's AI capabilities — Wavo assistant, Creative Hub generation, AI-powered rules. For deep-dive: see PRD-20 (Wavo) + PRD-18 (Creative Hub).

What's Wavo?

Wavo is Wevion's in-app AI assistant. Access at /chat. Answers questions about your data, takes actions (with approval), generates creatives via Creative Hub tools.

See ai-101 meet Wavo.

Which AI models can I use?

10 verified models across 5 providers:

Provider

Models

Anthropic

Claude Opus 4.6, Sonnet 4.5 (DEFAULT), Haiku 4.5

Google

Gemini 2.5 Pro, Gemini 2.5 Flash

DeepSeek

deepseek-chat, deepseek-reasoner (no tool support)

Moonshot

Kimi K2

OpenAI

GPT-4.1, GPT-4.1 mini

Switch per session OR set workspace default via /settings → AI preferences.

Can Wavo speak back via voice (TTS)?

No. Voice input via Whisper transcription only (25 MB max). Wavo doesn't generate audio output.

For voice-over generation (audio output for ads): use Creative Hub TTS — see ch-116.

Is there a mobile app for Wavo?

No native mobile app. Use:

  • Mobile browser (responsive UI)

  • PWA push notifications via VAPID

  • Telegram bot for real-time alerts (ntf-104)

How much do AI credits cost?

1 credit = $0.001 with MARKUP 1.1× applied on top of base provider costs.

Per-message cost varies by model + tokens. View per-message breakdown in /settings/team/billing.

See ai-109 credits + limits.

Can Wavo take actions on my account?

Yes — with explicit approval for HIGH / MEDIUM risk actions:

Risk

Tools

HIGH

delete_automation_rule, duplicate_entity, update_entity_budget

MEDIUM

create_automation_rule, toggle rules, toggle_entity_status

LOW

create_campaign_draft (no approval — drafts safe)

Approval card expires after 30 minutes if not confirmed.

See ai-105 actions.

Can I prevent Wavo from taking certain actions?

Yes — via role-based access control (RBAC). Viewers can't trigger write tools. Specific tools have role requirements.

Workspace-level overrides (which tools available per role): advanced configuration — contact admin.

Does Wavo see my data?

Yes — your team's data only (strictly team-isolated). Wavo cannot see other teams' data.

See ai-104 data grounded.

Are AI conversations private?

Yes — team-scoped. Sent to model providers per their enterprise terms. Most providers do NOT train on customer data (verify your enterprise agreement).

See ai-110 privacy + data usage.

Can Wavo generate images / videos?

Yes — via Creative Hub tools that Wavo can invoke:

  • Images: 3 providers (flux_2_pro, gpt_image_1_5, seedream_4_5)

  • Videos: 5 providers (luma_ray3, runway_gen4, kling_2_6, kling_3_0, veo_3_1)

  • Avatars: HeyGen (UGC-style videos)

  • TTS: ElevenLabs + OpenAI TTS

  • Compositing: Creatomate (assemble video + audio + text + transitions)

See ch-111 AI generation overview.

What languages does AI support?

  • Wavo: auto-detects user language; responds in same. Configurable via preferred_language in user_ai_preferences.

  • Whisper transcription: BCP-47 codes (en, it, es, fr, de, etc.); default it.

  • Creative AI: multilingual TTS (ElevenLabs) + image / video generation (provider-specific).

Does Wavo remember things across sessions?

Yes — via project memory facts:

  • Per-project: max 20 facts (MAX_PROJECT_MEMORY_FACTS)

  • Stored in project_memory_fact table

  • Opt-out: /settings → AI preferences → Memory toggle

See ai-106 projects + sessions.

Does Wavo hallucinate?

Wavo's system prompt forbids fabricating metrics. It cites the tool that produced each number.

If you suspect hallucination:

  • Ask "cite your source" — Wavo should reference the tool

  • Switch to a more capable model (Opus, Sonnet, GPT-4.1, Gemini Pro)

  • Give thumbs-down feedback to help improve

See ai-104 data grounded.

How fast is Wavo?

Depends on model + complexity:

  • Quick queries (no tools): 2-5 sec

  • Single tool call: 5-15 sec

  • Multi-tool chain (up to MAX_TOOL_ROUNDS=5): 30-90 sec

  • Creative generation (async): minutes (job polling pattern)

Faster models (Haiku, Flash, GPT-4.1 mini): noticeably quicker for short queries.

What if I want to disable AI features entirely?

Workspace-level: admin can disable AI features for the workspace (advanced config — contact support).

Per-user: opt out of memory + don't use chat. AI generation tools (Creative Hub) can be limited per role.

Can I integrate my own AI model?

No — currently fixed to the 10 verified models across 5 providers. Custom model integration not exposed.

For specific use cases (private models, custom fine-tuning): enterprise tier may include options. Contact sales.

Can Wavo connect to external knowledge (web, my docs)?

  • Web search: yes (web_search tool)

  • Your uploaded docs: yes (per-session file upload + project files — see ai-106)

  • Other knowledge bases: limited (no external KB integration beyond Wevion docs)

Does Wavo learn from feedback?

Yes — feedback aggregated for prompt + model tuning. Per-message thumbs up/down via PATCH /api/v1/chat/messages/:messageId/feedback. See ai-111 feedback + improve.

Can I see my AI usage history?

Yes:

  • Per session: messages persisted in chat_session

  • Total usage: /settings/team/billing → Recent transactions (credit charges)

  • Specific message details: chat_message includes token usage + cost

What's the AI Action Audit Trail?

Every approval-mode action by Wavo logs to ai_action_log:

  • Status (PENDING → SUCCESS / FAILURE / EXPIRED / DENIED)

  • Payload hash (SHA-256)

  • Idempotency key

  • Confirmed by + at

  • Executed at

Visible via /api/v1/audit-logs?resource_type=ai_action.

Related