Current

Archive and ingestion

Export multiple GitHub repositories with Git history, collaboration metadata, raw provenance, normalized records, attachments, integrity, and completeness.

Semantic indexing

Begin with TypeScript and JavaScript symbols, definitions, references, callers, callees, imports, inheritance, and test relationships.

Symbol lineage

Map symbols through unchanged, modified, renamed, moved, signature-changed, extracted, split, merged, deleted, ambiguous, and unresolved states.

Context packets

Return bounded, cited current structure, history, risk, planned work, and active work for a symbol, task, test, or proposed change.

Phase 2

Backlog synchronization

Attach tasks to durable semantic anchors, re-resolve them after change, and explain which assumptions require revalidation.

Phase 3

Change-risk and regression memory

Combine churn, co-change, reverts, follow-up fixes, CI, incidents, and explicit evidence without overstating causality.

Phase 4

Multi-agent coordination

Track time-bounded semantic claims and detect symbol, dependency, interface, schema, test, and contract overlap before integration.

Phase 4+

Decision memory and ADRs

Generate cited, human-reviewable decision drafts from commits, PRs, issues, reviews, transcripts, reverts, and outcomes.

Phase 1 vertical

Prove useful TypeScript/JavaScript context before broadening scope.

local repositories and KodeAtlas archives as inputs;
functions, methods, classes, interfaces, types, modules, and tests;
definitions, references, callers, callees, imports, inheritance, and test relationships;
commit, PR, and issue history attached to symbols;
first lineage states for ordinary edits, moves, renames, signature changes, and deletion;
MCP and CLI tools for context, history, impact, and anchor resolution;
a local inspector for trust and debugging.

Deliberate non-goals

Do not become a generic code-search or agent platform.

do not begin with every language;
do not treat line numbers as permanent identity;
do not infer certainty from temporal proximity;
do not hard-lock whole files by default;
do not replace GitHub, issue trackers, CI, or coding agents;
do not publish authoritative ADRs without human review;
do not depend exclusively on embeddings;
do not hide partial evidence.

Take the next step

Help test the layer that comes after archival.

The first context vertical should be judged by whether it materially improves an agent’s implementation plan—not by the size of its graph.

Discuss early access Read the roadmap

One evidence foundation. Progressive intelligence layers.