Skip to content

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

17 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Undes

npm Local-first BYOK Pro

Evidence-Checked AI Code

Ündes helps developers know what is safe to merge before trusting AI-generated code, patches, refactors, or architecture decisions.

It proposes a solution or code candidate, then checks it against visible evidence, files inspected, assumptions, what could not be proven, critique, open risks, and a conservative trust verdict.

npm install -g @undes.ai/cli
cd your-project
undes init
undes run --prompt="Review this change for risky assumptions, missing tests, and unsafe architectural drift."

What you get is not just a chat answer. Undes separates the candidate answer from the evidence and trust boundary around it.

Verdict: Needs review

Evidence:
- The validator rejects expired tokens but does not prove audience validation.
- Existing tests cover malformed tokens, but not cross-tenant tokens.

Rejected hypothesis:
- This is not explained by clock skew; expiry checks are already deterministic.

Open checks:
- Add an audience/tenant test before treating the change as safe to merge.

One prompt -> one evidence-checked candidate with trust boundaries. AI can generate the answer; Undes helps you decide whether it is safe to merge.

Why

AI made code generation cheap. The expensive part is trust:

  • What codebase evidence supports the answer?
  • Which assumptions did the model make, and which hypotheses did it reject?
  • Which risks are still open before merge?
  • Is the output a reviewable engineering artifact, or just a confident answer?

Undes makes that trust layer explicit.

Editions

Undes editions share the same core idea: AI generates an engineering answer, and Undes makes the trust boundary explicit before you rely on it. The editions differ by workflow, packaging, and support.

Edition For What it provides Status
Community Individual developers evaluating Undes or using it independently. Local-first CLI, BYOK model access, one prompt -> one evidence-checked candidate with trust boundaries, and operator-facing output. Available as @undes.ai/cli
Pro Professionals using the paid local workflow. License-gated local CLI, Pro terminal UI, local history, and native verification package. See undes.app/pricing
Team / Enterprise Organizations that need a broader commercial arrangement. Handled through direct discussion. No public package, install path, hosted workflow, or deployment promise is committed in these docs. Not self-service

When Undes Helps

Use Undes when a plain AI answer is not enough and you need a reviewable decision trail:

  • code review where the risk is hidden in assumptions, not syntax;
  • bug investigation where several hypotheses look plausible;
  • architecture or migration decisions that need explicit open checks;
  • validation, auth, billing, or data-flow changes where a confident answer is not sufficient;
  • local BYOK workflows where repository content should not go to Undes servers.

Repository Structure

  • community/ — Community install notes, command reference, license scope, security notes, and release changelog.
  • pro/ — Pro command reference, external CLI setup, and public setup notes.
  • docs/ — shared product documentation for the core Undes workflow.
  • articles/ — public articles and long-form product explanation.
  • examples/ — sanitized examples of generated and trust artifacts.

Start with:

What a run produces

An evidence-checked engineering artifact:

  • the proposed implementation or diagnostic answer;
  • the codebase evidence supporting it;
  • assumptions and open checks;
  • rejected hypotheses;
  • risk notes;
  • a trust verdict.

See the demo walkthrough and the sanitized artifact example for the shape of the output.

About this repository

This is the public documentation, examples, and feedback surface for Undes. It is not the product source — Community is installed as a CLI, and the Pro and Enterprise editions are delivered separately.

Feedback

Open an issue describing the task you tried and which trust signal you would have needed before deciding to merge, defer, or reject the result.

Releases

Packages

Contributors