Khala

Grounded knowledge

Every answer
cites a source.

Khala answers questions about your code, docs, and services using only evidence it can cite. When nothing supports an answer, it says so instead of making one up.

Get startedRead the philosophy →

Start here

01

Getting Started

Install it and get your first grounded answer in a few minutes.

→02

Philosophy

The idea behind Khala, and why grounding matters.

→03

Ecosystem

How the tools connect through one shared link.

→

Two ways AI work goes wrong

01

Confident wrong answers

Models state stale or incorrect facts with full confidence. Nexus and Archon answer only from sources they can cite, so every claim comes with its evidence.

02

Unreviewed AI output

AI-written code gets approved without a real read. Arbiter makes review an explicit, recorded gate before code is written.

One substrate, four kinds of information

01

What the org knows

One warehouse, two doors — humans (web) and agents (MCP) read the same governed corpus: same approvals, same current version, same citations.

Nexus

02

Why it was built

A flight recorder for design decisions — the choices agents make by the hundred, recorded at zero cost and approved as a named human’s accountable act.

Arbiter

03

What the system is doing

Judgment context, not another dashboard — telemetry joined with approved knowledge into evidence for review and troubleshooting.

Observer · Nexus

04

Who still understands it

A cognitive-debt ledger — the warehouse is the denominator, vouches the numerator; the gap becomes a number the team can repay.

Adept

Three debts it pays down

01

Technical debt

Code and artifacts pile up faster than anyone maintains them.

Serviced by Probe · Observer

02

Intent debt

The reasons behind a decision, and its trade-offs, get lost.

Serviced by Arbiter

03

Cognitive debt

No one fully understands the system the team ships. Closing that gap is the point of Khala: you can still read and trust what the AI built. Adept turns the gap into a number — vouch coverage — so repayment can be planned.

Measured and repaid via Adept

The tools

RAG · GraphRAG

Nexus

Answers from your docs and telemetry, each with a citation.

→Domain truth

Archon

Authoritative answers about your domain’s rules and constants.

→Code review

Observer

Reviews PRs for scope, API contracts, and cohesion.

→Spec / ADR gate

Arbiter

Makes spec and ADR approval a recorded gate before coding.

→Mutation testing

Probe

Mutation testing that finds gaps a green suite misses.

→Cognitive-debt ledger

Adept

Measures whether a named human can still vouch for what shipped.

→

Resources

Contributing →Source & releases →