Technical

How, exactly.

One hub, thin clients, typed tools, field-level data governance and a paper trail on every run. This page is for the person who has to say yes to the architecture.

Hub and client

One server runs the AI. Your Odoo stays thin.

The 360 AI server is itself an Odoo instance. It holds the providers, prompts, agents and logs, and manages every connected customer environment. The client module on your Odoo is deliberately AI-free: it calls the hub and renders results.

Data governance

The model sees what you allow. Field by field.

Every prompt is bound to a schema: an explicit selection of which fields, relations and context (message history, activities, attachments) get serialized to the model. Not "the record", but the fields you chose, with hidden fields that never leave Odoo.

Schemas

Field-level allowlists per model, explicit or additive, with hidden fields and controlled recursion into relations. Privacy and token cost, managed in one place.

Token metering

Usage tracked per provider connection, with a configurable backup provider when the primary fails.

Logged runs

Action logs on every AI run, timed agent executions, Logfire instrumentation. When someone asks what the AI did, you can show them.

Human-in-the-loop

Supervised agents defer writes to an approval queue. AI decision cards land in a systray inbox on the client for review.

Evals with a judge

Datasets (including synthetic), LLM-as-judge scoring from 0 to 10 with reasoning, side-by-side provider comparison. Prompt changes get graded before they ship.

Functional tests

UI-callable tests exercise the live LLM integrations as background jobs, so upgrades don't silently break AI features.

Protocols

Three MCP surfaces, one standard.

360 AI doesn't just expose MCP, it speaks it in all three directions.

Odoo as MCP server

The /ai360mcp endpoint serves nine typed tools, with OAuth 2.0 sign-in (PKCE, dynamic client registration, refresh tokens) or per-user bearer keys. Calls run as the authenticated user.

The hub as MCP relay

The /x360mcp relay routes MCP tool calls to any connected environment, behind environment + user keys, model and method allowlists. The full story is on the MCP page.

Agents as MCP clients

Register any external MCP server (SSE or streamable HTTP) and its tools join the agent toolset, next to the native Odoo tools.

MCP Inspector connected to the x360 Remote Odoo MCP server, listing the seven x360 tools
MCP Inspector against the live relay: connected, seven tools listed, version 19.0.1.0.0.
JSON-RPC 2.0 MCP 2024-11-05 → 2025-06-18 Streamable HTTP OAuth 2.0 · PKCE REST · FastAPI XML-RPC
The stack, honestly

Boring where it should be.

No exotic runtime. Odoo modules, a job queue, pydantic-ai for the agents, and code you can read.

55
Modules
install what you need, skip the rest
4
Odoo versions
clients on 16, 17, 18 and 19
2
Provider backends
AWS Bedrock + any via Instructor
0
License surprises
LGPL-3.0, public repo

Bring your architect.
We like the hard questions.

360 ERP runs the platform and implements it. Ask for a technical deep-dive session and bring the person who says no.

Talk to 360 ERP