Software and coding
Your team’s own coding model.
Coding models that learn your own codebase, conventions and reviews.
StatusOne of our first clients works here
Two comments: the new endpoint skips the auth middleware your rules require 1, and the query should use the shared pagination helper 2.
- 1Engineering rules, security
- 2Pagination helper, shared library
Illustrative
- Tasks to start with. A Test Lab takes one.
- 3
- Its Formula
- LLM
- To roll back to the last release, by design
- 60 s
- Eigna Cloud. Data stays in the region you choose.
- Cloud
The work that cannot leave
Your code is your product. Coding tools already learn from what developers accept. Here, the model that learns is yours.
- Source code
- Reviews and changes
- Internal docs
- Incident notes
Where it runs, for software
Eigna Cloud
Data stays in the region you choose.
A dedicated GPU in the region you choose, with no token bill. Or Eigna Local, on your own machines.
Your data sits on your own volumes in our GPU region, with keys you hold.
The first expert
Three tasks to start with.
Reviews and changes your engineers approved.
Code review
Against your own rules and past reviews.
Two comments: the new endpoint skips the auth middleware your rules require 1, and the query should use the shared pagination helper 2.
- 1Engineering rules, security
- 2Pagination helper, shared library
Learns from · Reviews your engineers approved.
Drafts in your style
Following your conventions and libraries.
A draft that follows your migration conventions 1 and the naming in the billing module 2.
- 1Migrations guide
- 2Billing module
Learns from · Changes your engineers merged.
Answers from your docs
With the file and the line it came from.
Use the admin task in the operations runbook 1. The code lives in the keys service 2.
- 1Operations runbook, keys
- 2keys/rotate.ts, line 41
Learns from · Answers your engineers confirmed.
The proof we will measure
A metric, a baseline and a sealed test, agreed before we start.
The target is set with you before the Test Lab starts. If the expert does not beat your baseline, you keep the report.
17.8%
Our first training run cut forecast error 17.8% against its base model. It is not yet ahead of a simple last value baseline, and we publish that too.
Rules in plain words
What the rules ask, and what every release gives you.
- 01
Your code trains your model
Your repositories and reviews train only your model. They never move between clients.
- 02
Licences checked
Open base models are examples; every licence is checked before use.
- 03
A cost you can plan
A fixed running cost instead of a bill per token.
Software and coding
Your code trains your model.
Your repositories and reviews train only your model, at a fixed running cost instead of a bill per token.
Software: questions and answers
Do we have to drop our current assistant?
No. Keep it for broad work. Own the model for your codebase, your conventions and your reviews.
What does it learn from?
Changes and reviews your engineers approved. A suggestion that was rejected is not training data.
Where does our code go?
With Eigna Cloud, your own volumes in the region you choose, with keys you hold. With Eigna Local, it stays on your machines.
Test Lab
Request a Test Lab for software.
One task, your data, a sealed setup, about four weeks. If the expert beats your baseline, the test set becomes your first release gate. If not, you keep the report.
- RouteDirect, and through Eigna Ventures for AI-first startups.
- Watch forCoding assistants are good at broad work. Owning pays on your own codebase and your own rules.
- ReplyWithin two working days, from a founder.
