anora.labs
The platform

One platform to deploy AI on top of the systems you already run.

Data integration, context, orchestration and security. Deployed solution by solution, in your infrastructure.

  • Agents and operator apps for each process. The agent prepares, the person decides.

    • Operator apps for the people who decide
    • Everything exposed through API and MCP
    • Simple automations configured by your own team
  • Every agent runs with its own permissions, approval policies and a full log.

    Permissions
    Each agent sees only what its role needs.
    Approval
    Human checks wherever you want them.
    Logging
    What it did, with which data and context.
    Evaluation
    Tested on real cases before production.
  • Whether an agent works depends on the data and context it can reach. This layer holds both.

    Structured · Data warehouse
    Data from every system, one definition per metric.
    Unstructured · Context graph
    Decisions, preferences and rules from Excels, PDFs and people.
    Inside the context layer ↓
  • Connectors to the systems you already run. No migration: the ERP stays the system of record.

    • Reads from reporting or a replica
    • Writes back only through the channels the ERP exposes, with approval
    • ERP, MES and CMMS, CRM, Excels and email
In practice

A delivery slips five days. The planner sees three options, one recommended, and applies it in a click.

The sourcing agent sends the RFQs on its own. A person approves the order. Every run is logged.

"Under 500 units per batch, sales approves" is stored once. Every agent and every person sees the same rule.

SAP orders, supplier emails and the planning Excel land in one place, without touching the ERP.

Inside the context layer

Most processes need more than data: past examples and rules learned over the years. Agents need the same.

Structured data · Data warehouse

Each metric is defined once, versioned, and read the same way by reporting and by agents.

metrics/otif.sqlv14 · current
select week,
  count_if(on_time and in_full) / count(*) as otif
from clean_orders
group by week

One file per metric, one rule. Every change versioned.

Unstructured data · Context graph

Entities, documents and rules, linked. Agents and people get the answer with the rule and its source attached.

Archive
Entities, relations and rules extracted from emails, PDFs and Excels.
Team
Interviews with the people who know each process best.
Every run
Corrections, approvals and exceptions, captured for next time.
Rule · Minimum per batchNode
Rule
Under 500 units per batch, sales approves
Applies to
References A-100 to A-140
Source
Purchasing email · March 2024
Status
In force · reviewed Feb 2026
Links to
12 references · 3 open orders · Line 3
Deployment

Runs where you decide, with the models you choose.

Where it runs

In your infrastructure, or in our cloud if that suits you better.

Any model

Each task uses the right model for price, quality or speed. Including local models.

Your data

Data, definitions and context stay on your side.

Open

Everything is exposed through API and MCP.

Some of the systems we connect to

See which process to start with.

Book a scoping call