Product inventory

Capabilities

The shop window stays light on purpose. This page is the deeper list: what the AI suite can do today — connectivity, knowledge, governance, experience and platform — without turning the pitch into a feature dump.

Connectivity

Talk to Odoo — and beyond

MCP server

Expose your Odoo instance to MCP clients (Cursor, Claude Code and others) under the same permission model.

OpenAPI servers

External MCP servers

Connect outbound to other MCP endpoints (SSE or stdio), with trust configured per server.

Interlinked Odoo instances

Link another Odoo by exposing it as MCP or OpenAPI — no proprietary federation lock-in.

Controlled web egress

Optional HTTP fetch to allowed domains only, with time-bounded validity.

Knowledge

How the company teaches the agent

Agents, roles & bundles

Each agent runs with a role, a provider chain and exactly one context bundle — clear composition, not a soup of prompts.

Contexts

Declarative company knowledge (rules, jargon, policies) injected by locale.

Skills

Reusable procedures the operator can invoke; listed and retrieved like tools, not buried in chat history.

Import / export of knowledge

Move bundles, contexts, skills, providers, servers and whitelist policies between instances as ZIP/JSON packages.

Sandboxed read (RelaxiCode)

Bounded Python reads for exploration — with rollback, without direct writes.

Governance

Propose ≠ execute

Write governance (Safe Plan)

The model proposes; a human confirms; the server executes. Prompt text alone cannot unlock writes.

Approval history

Pending, confirmed and cancelled operations stay visible — who authorized what, and when.

Low-level AI operation log

Audit trail of origin, client, steps, failover and correlation for support and compliance.

URL whitelist

Instance policy for external domains: whitelist-only or open, importable as JSON.

Capability groups

Separate rights for writing, external URLs, external APIs and administration — trusts do not collapse into one superuser.

Experience

Chatboo inside the ERP

In-Odoo chat

Conversational assistant that runs in your instance, not a side SaaS.

In-flow AI

Chat on top of your work session, toggled from the tray icon. It does not interrupt or replace the screen — show, ask, hide.

Screen context

Ask about the active record or list underneath the chat without re-explaining where you are.

Streaming answers

SSE streaming so long answers feel live, not frozen.

Answer modes

Default layouts: paragraph, table→chart, chart→table, and dashboard — so a list stays a list unless you ask for a chart.

Session history

Named, reloadable conversations per user — not ephemeral prompts that vanish.

Platform

Built to last in your clients' installations

PAAP

Open Patanegra Application Agent Protocol: roles, write auth, host profile and presentation on top of MCP.

Open LLM providers

OpenAI-compatible stacks, Anthropic, local models (Ollama / LM Studio…) with per-agent failover.

Odoo 14 and 19

One product line, the same module in every client's Odoo.

On-prem sovereignty

No Enterprise IAP dependency; Apache-2.0 modules; credentials stay in your environment.

For module developers

You own the domain problem — AI plumbing is already solved

Agentic protocol (PAAP)

Providers, APIs, history, sessions and write authorization are shared infrastructure — not something each module reinvents.

Focus on your domain

You solve the functional problem. The engine and Chatboo already give the agent runtime.

Optional contexts & skills

When useful: declare shareable domain knowledge (contexts) and agent-owned procedures (skills). Not mandatory boilerplate.

Composition, not prompt soup

Bundles and roles assemble what each agent needs without fighting other modules’ APIs or providers.

Import / export of artefacts

Package contexts, skills and related config between instances when you deliver consulting knowledge as files.

Apache-compatible stack

Reference AI modules are Apache-2.0. Your original module/agent code may use another license; Apache parts keep Apache terms — see Licenses.

Honest limits

Where a generalist AI does better

Credibility matters. A chat in Claude, ChatGPT or similar may win on pure generative fireworks. That is fine — and it clarifies what we sell: integration into the ERP, not a prettier chatbot.

Output variety

Generation is solid for ERP work, but not as wide as a frontier generalist. We do not ship endless exotic chart styles or open-ended creative formats.

Open brainstorming

For free-form ideation, long essays or tasks far from your data, a generalist chat is often the better desk tool.

What we double down on

Permissions, screen context, governed writes and company knowledge inside Odoo. That is the heavy lift — the model underneath stays replaceable.

On providers

Depending on a good model is common to every AI product — it is not a flaw of this one. We recommend connecting through an LLM router (a gateway that lets you pick models by quality and price, and switch without rewiring the ERP). A modest monthly budget often covers a large volume of solid queries without locking you to a single vendor.

What this page is not

It is not a roadmap catalogue. Items here are available in the current modules (AI Engine + Chatboo and companions). Some ideas — answer modes, in-flow AI, charts — will need screenshots later; vertical packs and demo agents still in construction belong elsewhere when they ship.

Want the story, not the inventory?

Start with Integrated AI or the Consultant page. Come back here when you need the full map.

Consultant → Integrated AI