Rationale is distributed
Important intent lives across Git, pull requests, issues, reviews, tests, incidents, documents, and agent sessions.
Temporal code intelligence · Archive foundation in place · Context platform in development
KodeAtlas gives developers and AI agents the historical, architectural, planned-work, and active-work context surrounding meaningful parts of a codebase—backed by the evidence that produced the current system.
KodeAtlas CLI v0.2.0 currently provides deterministic, resumable, verifiable Git and GitHub archival. Symbol lineage, context packets, backlog awareness, and multi-agent coordination are later product phases.
The context gap
Present-state search can show dependencies. It cannot reliably explain which alternatives failed, which assumptions remain active, or which planned work is about to collide with the same contract.
Important intent lives across Git, pull requests, issues, reviews, tests, incidents, documents, and agent sessions.
Tasks and documentation become stale when symbols move, split, merge, or change signatures.
Prior regressions, reverts, and historical co-change patterns rarely reach the agent before it edits.
Two agents can change the same API or contract from different files without touching the same lines.
Available today · KodeAtlas CLI v0.2.0
The current KodeAtlas CLI exports related repositories into one local evidence archive. It preserves source history and collaboration history together, keeps raw evidence alongside normalized records, records completeness and provenance, and verifies the archive with durable integrity manifests.
kodeatlas initkodeatlas exportkodeatlas updatekodeatlas verifykodeatlas inspectOne evidence foundation. Four progressive layers.
Current and future capabilities remain visibly separated so the website explains the direction without presenting roadmap work as shipped product.
Preserve Git and GitHub evidence with provenance, completeness, resumability, and verification.
Index TypeScript/JavaScript symbols, track lineage, and deliver cited context and impact packets through CLI and MCP.
Attach backlog items to durable semantic anchors and identify work invalidated by code or contract changes.
Let agents declare time-bounded semantic intent and warn before contract overlap becomes a merge or runtime failure.
The intended interaction · In development
A context packet is designed to be small enough for an agent to use, specific enough to change the plan, and cited enough for a person to trust.
Calculates subtotal, tax adjustments, credits, and final total.
Introduced in PR #142; tax logic moved in PR #191; rounding strategy corrected in PR #247.
High churn; two prior regression associations; one open task targets the return contract.
Another agent is modifying InvoiceSerializer. Semantic conflict: medium.
Every statement links to a commit, PR, issue, test, task, decision, or active claim.
At the moment of change
Retrieve current dependencies, tests, history, earlier regressions, and related decisions.
Resolve planned work to the current symbol after a rename, move, extraction, or signature change.
Explain which task assumptions changed after a merge instead of silently rewriting the task.
Build an evidence-scored timeline of relevant changes, reverts, failures, fixes, and coupled code.
Warn when agents changing different files are altering the same API, schema, test boundary, or downstream contract.
Generate a cited ADR draft from the discussion and outcomes that produced the current architecture.
Evidence before narrative
KodeAtlas separates exact facts, deterministic derivations, and uncertain inferences. Symbol lineage can be ambiguous. Regression proximity is not causality. A context packet may be incomplete because a private fork, transcript, incident, or CI record is unavailable.
The product direction requires those limits to remain visible, preserves competing candidates, and treats human corrections as new evidence instead of erasing earlier results.
Read the trust modelSensitive software memory
The first context and anchor workflows are intended to run against local repositories and locally controlled archives. Team and enterprise modes can later add shared services, access control, retention, audit, model-provider policy, and air-gapped deployment.
Review the security directionThe Kalu Kode product family
Four distinct products share one evidence-first philosophy and remain independently useful.
Preserves the observable truth of deployed software as replayable evidence.
Early accessRecoverRecovers maintainable source from software that still runs.
Selected assessmentsPreserves why the code became what it is and delivers that context where change begins.
You are hereControls what AI-authored changes reach the repository.
Private alphaEarly context vertical
We are looking for technically engaged TypeScript/JavaScript teams willing to evaluate context packets, durable anchors, and agent workflows against real repositories.
Take the next step
Explore the verified archive foundation now, or discuss the first TypeScript/JavaScript context vertical being built on top of it.