Inline completion
Context-aware suggestions as developers type, in the editor they already use.
PAI-CODE
Private Coding Assistant
AI pair programming that keeps your source code in-house.
Give your developers the productivity of an AI coding assistant without sending a line of source code outside the company. Completion, chat, code review and agentic tasks run inside VS Code, JetBrains IDEs and the terminal, powered by open-weight coding models on your appliance and governed by your own access rules.
- Completion, chat and review in the IDE
- Repo-aware answers from your own code
- Source code never reaches a public model
01The challenge
Source code is your most valuable, least protected asset.
Cloud coding assistants send code context to external servers. For proprietary systems, integrations and customer projects, that is often not acceptable. A private coding assistant keeps the benefit and removes the exposure.
02Capabilities
What it does.
Repo-aware chat
Ask how a module works or where a function is used. Answers draw on an index of your own repositories.
Code review and summaries
Explain changes, suggest improvements and draft pull-request descriptions.
Tests and documentation
Generate unit tests, docstrings and README updates for existing code.
Agentic tasks with approval
Multi-step changes planned and executed in the IDE or terminal, with a developer approving each step.
Governed access
Single sign-on, per-team model access, usage reporting and a full audit trail.
03How it works
From question to answer, inside your walls.
Open-weight coding models served through a private, OpenAI-compatible gateway, with open-source IDE extensions such as Continue and Cline for VS Code and JetBrains, plus terminal clients for agentic work.
- 01IDE and terminalVS Code, JetBrains, CLI
- 02Private gatewaySSO, policy, audit
- 03Repository indexStays on-premise
- 04Coding modelServed on the appliance
- 05SuggestionDeveloper reviews and accepts
- 01
Install the IDE extension
Open-source extensions for VS Code and JetBrains, or a terminal client, point at your private endpoint.
- 02
Sign in
Developers authenticate with single sign-on; access follows team and project rules.
- 03
Index your repositories
Selected repositories are indexed on the appliance for repo-aware answers.
- 04
Code with AI
Completion, chat, review and agentic tasks are served by coding models on your hardware.
- 05
Measure and govern
Usage and adoption are reported per team, with every request auditable.
04A day with the assistant
What developers actually ask it.
The assistant is most valuable on your own code, where public tools cannot safely go.
Explain
Walk me through how the order allocation service reserves stock, and where it can fail.
Test
Write unit tests for the costing module, covering currency rounding edge cases.
Review
Review this pull request for security issues and summarise the change for the release notes.
Modernise
Document this legacy ERP customisation and propose a refactor into smaller functions.
05Use cases
Where it pays off first.
Onboarding to large codebases
New developers ask questions of the code instead of waiting for a senior engineer.
Legacy modernisation
Understand, document and refactor older systems and ERP customisations safely.
ERP, MES and integration work
Faster development of connectors and internal tools around core business systems.
Test coverage
Raise coverage on critical modules without slowing delivery.
Secure code review
Spot risky patterns before they reach production, on-premise.
Documentation that stays current
Generate and refresh technical documentation from the code itself.
06Security and governance
Controls built in, not bolted on.
Every Logic Sonata system runs on hardware you control, with access, audit and data boundaries designed in from the start.
- Code and prompts processed only on your infrastructure
- Repository index stored on the appliance
- Per-team access to models and repositories
- Agentic actions require developer approval
- Full audit trail of requests
- Model choice without vendor lock-in
07FAQ
Questions about PAI-CODE.
How does it compare with cloud coding assistants?
For everyday completion, explanation and test writing, current open coding models perform well. The difference is control: your code never leaves the company. We benchmark against your own tasks during the pilot.
Which IDEs are supported?
VS Code and JetBrains IDEs through open-source extensions, plus terminal-based clients for agentic workflows.
Can we choose the model?
Yes. The platform is model-independent, so we select coding models to suit your languages, hardware and licence requirements, and can change them later.
08More from Logic Sonata
Other private AI systems.
Private Knowledge Assistant
A governed assistant that answers questions from your SOPs, specifications, policies and reports. Every answer cites its source and every user sees only what their role allows.
Private Vision Intelligence
Real-time analysis of live camera feeds by a vision-language model on your appliance. Ask what is happening on a line, at a dock or at an entrance, without video leaving your premises.
Private AI Agent Platform
Self-hosted AI agents built on Hermes Agent or OpenClaw that research, write and act across your tools, running on private models inside a sandbox with approval for anything that matters.
Private Image Generation Studio
A ready-to-use image generation studio built on ComfyUI, preinstalled with curated models and templates, so designers create prints, concepts and campaign visuals without their work leaving the building.
Get started
See PAI-CODE on your own data.
Book a 30-minute consultation. We will show the system working and scope a pilot around your use case.