Context gets lost
Beyond a handful of files, the assistant no longer sees the dependencies between modules or the history behind design choices.
Junyr.eu
AI that understands your whole codebase, not just the open file.
Our harness puts coding agents to work on large, long-lived repositories: it gives them the right context, has them plan, edit and test, then hands every change to your developers for review. Bring your own LLM, with a full on-premises option: your code never leaves your walls.
The problem
AI assistants stall on large codebases.
It makes things up
Functions that don't exist, guessed APIs, internal conventions ignored: the code fails to build or breaks something elsewhere.
Your code leaves the building
Sending your software assets to a public AI service raises real intellectual-property and confidentiality questions.
How it works
A harness, not just an assistant.
The model proposes, the harness decides what's allowed. Every step is logged, and your developers stay in control of what goes into the repository.
- 1
Map the code
The repository is indexed into a graph: modules, dependencies, entry points, tests, documentation. Agents query this map instead of reading code blindly.
- 2
Supply the right context
For each task, the harness selects the files, interfaces and conventions that matter, within the model's context budget.
- 3
Plan, edit, test
The agent writes a plan, applies the changes, runs the tests and iterates until they pass, with minimal permissions and a full log.
- 4
Human review
Every change arrives as a reviewable proposal. Nothing is merged without a developer's approval.
Your code stays with you
Sovereign by design.
Bring your own LLMAny OpenAI-compatible model, including open-weight models served with Ollama or vLLM. No model imposed.
Full on-premises optionHarness, index and model on your own servers: the code never leaves your infrastructure.
Three confidentiality levelsYou decide what may be processed by which model, repository by repository.
Full traceabilityEvery agent action is logged: who, what, why, and with which test result.
Everything is yoursConfigurations, agent mandates, prompts and skills are handed over to you.
Transparent costsToken budget estimated and tracked, with no surprise variable bills.
Questions about hosting, models or confidentiality? Read the Junyr FAQ →
Use cases
Where the harness makes the difference.
Legacy modernisation
Migrate a framework, a language version or a database, module by module, backed by tests.
Cross-cutting refactors
Rename, split and deduplicate across hundreds of files without breaking call sites.
Test generation
Cover untested code before touching it, so every later change is safer.
Living documentation
Describe the real architecture from the code itself, and keep it up to date.
Developer onboarding
New joiners query the codebase and understand in days what used to take weeks.
Fixes and maintenance
Reproduce a bug, locate the cause, propose a tested fix.
Getting started
From audit to rollout, in three steps.
30-minute audit
Your codebase, your hosting constraints, and the first use case to target.
Pilot on one repository
The harness installed on a real repository, metrics measured before and after, and a clear go/no-go decision.
Rollout and autonomy
Extension to your other repositories and team training through the Agentic Processes course.
FAQ
What engineering teams ask us.
Is our code sent to an AI provider?
Not unless you want it to be. In on-premises mode, the harness, the code index and the model all run on your infrastructure. You can also choose a European-hosted model for less sensitive repositories.
Which models can we use?
Any model exposing an OpenAI-compatible API, including open-weight models served with Ollama or vLLM. The model can be swapped at any time.
Can the agents change code without oversight?
No. Agents work with minimal permissions and propose their changes; a developer reviews and approves before anything is merged.
Do we need to replace our current tools?
No. The harness plugs into your existing version control and test suite. The exact scope is defined during the audit and the pilot.
How much does it cost?
The audit is free. The pilot is quoted according to repository size and hosting mode, with a token budget estimated upfront.
30 minutes to assess your codebase.
We'll look together at your repository, your confidentiality constraints and the first project where an AI coding harness can save your team time.