Attach unlimited alternate VRM outfits to a single character. Swap looks live during a session — uniform during demos, casual after hours, branded outfits for trade-show kiosks.

Outfit description per record — the agent reads it and stays in-character when the look changes.

Attach background images per avatar. Set the mood — office, warehouse, conference floor, anime-cafe.

Combined with Grok Imagine tools, the avatar can generate new backgrounds and image content on demand.

Upload .vrma motion files per avatar — built-in gestures include thinking, clapping, jump, peace_sign, spin, dance, look_around, sleepy, goodbye, shoot, squat, model_pose — and the agent picks the right one for the beat.

The agent calls set_emotion to blend VRM facial expressions in real time — happy, sad, surprised, relaxed, angry, neutral. The face matches the voice as the conversation unfolds.

Copy a voice ID from the Grok Console voice library, paste it into a Custom Voice record in Odoo, and assign it to any agent. Re-use the same voice across multiple agents.

Upload an audio sample directly from Odoo. The module submits the sample to xAI, tracks the state, and exposes the resulting voice ID for use on any agent. Subject to xAI custom voice region restrictions.

• search_read — records by domain
• read_group — aggregations and counts
• list_models — discover available models
• get_model_schema — field inspection

Always scoped by the user's Odoo ACL

• create_record — opt-in per agent
• update_record — opt-in per agent

Off by default — enable only on trusted agents

• find_action — locate a destination by free text
• open_record — jump to a record by model + id
• dispatch_action — open any list, kanban, form, pivot, graph, calendar or gantt, with filters / group-bys / search terms applied
• run_view_action — open client actions (Discuss, Knowledge, dashboards) and printable reports

• set_emotion — blend a facial expression
• play_gesture — trigger a VRMA clip
• change_outfit — swap to another VRM costume on the agent
• change_background — switch the avatar backdrop (preset or Imagine-generated)

• dom_snapshot — capture the page's interactive elements as a numbered accessibility-tree list
• dom_action — click, fill, select, press, hover or focus any element by ref (handles rich-text editors and Ctrl+S-style shortcuts)
• dom_extract — read an element's text or any attribute to verify state after an action

Used when the curated navigation tools can't reach a specific button or free-form field

• Web search — live search of the open web
• X search — live search of posts on X
• Code execution — sandboxed Python
• create_image — Grok Imagine generation, persisted as rexclaw.voice.imagine.image records
• edit_image — Grok Imagine edit of an attached image, same persistence

Voice Summarization Threshold (tokens) — the token count at which a voice session triggers a compaction pass. Lower it for tighter memory and lower per-turn cost; raise it for richer working context.

Default sits in a sensible middle range — tune per the verbosity of your conversations and your xAI budget.

Text Summarization Threshold (tokens) — the same dial, but for text sessions. Text turns are typically longer and image attachments add tokens, so this is a separate knob.

Useful when text mode is doing heavy lifting (image analysis, long prompts) and you want different memory tactics than voice. Note: text mode runs on the xAI Responses API, which has a different context window than the voice realtime API — so the right threshold here is usually higher than the voice one.

Recent turns to keep verbatim when summarizing — how many of the most recent turns survive a compaction pass verbatim. Older turns become a summary; these stay word-for-word so the agent has crisp recent context to anchor on.

Summarization Model — which xAI model runs the compaction. Defaults to grok-latest, an alias that always points at the latest Grok model.

Transcript Retention (days) — how long sessions (and their compacted summaries) are kept before a daily cron purges them. Long retention gives companions deep memory across weeks; short retention favours privacy and storage.

Transcript display limit (messages) — how many messages render in the live transcript pane. Independent of what the agent remembers — this is purely a UI cap to keep the side-panel quick to scroll.

xAI Endpoints & Models

Realtime, client secrets, responses, files, images, image edits URLs — plus separate Voice Model, Text Model, Summarization Model, Imagine Model. Override any of them when xAI ships a new tier.

Daily Per-User Caps

Voice Daily Token Cap (per user, estimate) + Text & Imagine Daily USD Cap (per user). Hard ceilings, enforced server-side. No surprise xAI bills.

File Expiry

File Expiry (seconds) — how long user-uploaded attachments in text mode live on the xAI Files API (Responses path). Capped at 30 days by xAI; defaults to that maximum. Not used by voice mode or for Imagine-generated images.

Default Agent

Default Agent — the agent every user starts with until they pick their own on the preferences tab.

All Odoo tool calls execute as the connected user — the agent inherits their record rules and ACLs. It cannot see, create or update anything they couldn't.

Only members of the Voice Companion User group can open a session or list agents shared with them. Admins manage agent ownership and sharing per record.

Session transcripts are kept for a configurable number of days. A daily cron purges anything older. Generated files have their own expiry.

Voice token and text/imagine USD caps are enforced per user, per day — no surprise xAI invoices.