AI-ops · Internal speed

Work with 1C in the language of business — without risking the core

The agent runs on Claude: developers prepare changes faster via prompts, plans and extensions, while managers read and update data only within their allowed scope. Separated roles, review flows, audit log throughout.

Book a free call What's included

The Problem

Sound familiar?

Every 1C change queues behind the dev team, requires manual requirements handoff and risks breaking the standard configuration.

Managers don't need access to everything — they need fast, safe actions inside their role.

Developers drown in lookalike tickets — reports, reference data, exports — and can't focus on harder work.

Companies want AI inside 1C but are afraid to let it near the core and critical data without control.

We get it

“We don't let AI change the core freely, we don't blur manager and developer roles, and we don't sell a magic button.”

Our track record

We build controlled acceleration over 1C: a permissioned AI layer where every action passes through roles, approval and an audit log.

How It Works

Three steps to get started

Map roles and limits

We describe roles, accessible objects and hard limits. We decide what managers can do and what only developers can do.

Connect 1C

We connect 1C via OData, HTTP and web-services and expose only the actions each user type actually needs.

Launch the flows

Developers: prompt → plan → extension → review → apply. Managers: prompt → allowed action → journal → control.

Deliverables

What's included

Concrete deliverables. Each item below ships as part of the initial engagement, not as a future upsell.

Permission-based AI layer over 1C built on Claude

Developer mode: prompts, change plans, extension prep, review-friendly diffs

Manager mode: read, search, create and update data within allowed objects

Separate roles and permissions for developers and managers

Action approval before write or publish

Change journal, request history and transparent audit

1C integration via approved tools: OData, HTTP services, extensions

MCP connections to external systems via allowlist/denylist

Plan review and permission flows in the Claude Code style

Training for developers and managers on the agent

Results

What you can expect

Routine changes and exports get done by developers an order of magnitude faster

Managers work with data in plain language, never touching the core

Every action goes through review, approval and the audit log

AI lives inside 1C, not next to it — one source of truth

The cost of waiting

If the business needs speed and IT needs safety, you don't need another chatbot. You need a managed AI layer over 1C.

FREE AI AGENT

FAQ

1C Agent — FAQ

30-minute discovery call. No pitch, no pressure. We'll build and deploy your AI agent for free.

Can managers actually change 1C data through AI?

Yes — but only within their role. If a sales manager can create a client and an order, the agent lets them do it via prompt. If they can't, the agent won't offer the action. This is AI through permissions, not AI instead of them.

How is the core configuration protected from AI?

AI never writes to the base configuration. All changes go through configuration extensions, plus plan review and developer approval before apply. The standard configuration stays intact and upgradeable.

What if the AI makes a mistake preparing a change?

That's why the dev flow is prompt → plan → extension → review → apply. The developer sees the plan and the diff before apply. Nothing lands in prod without explicit approval. This is the Claude Code pattern — accelerator, not autopilot.

Which 1C versions do you work with?

1C:Enterprise 8.3+, standard configurations (Accounting, Trade, KA, ERP) and custom ones. Integration via OData, HTTP services and web-services.

Where are prompts and context stored?

In your perimeter — your server or your cloud. Claude models are called via API on business tier, where data isn't used for training. For strict requirements an on-prem model is possible.