KODEATLASby Kalu Kode
Discuss early access

Product roadmap

Archive first. Context next. Coordination only after trust.

The roadmap starts with a narrow TypeScript/JavaScript vertical and grows only after symbol history and context prove useful.

Implemented

Phase 0 — Archive foundation

Stabilize the multi-repository archive CLI, import formats, schemas, integrity, provenance, normalized evidence, activity, and relationship indexes.

Acceptance outcome: Verified Git/GitHub archive with explicit completeness and immutable run receipts.

Next

Phase 1 — Symbol history and MCP

Index TypeScript/JavaScript symbols, connect them to commits, PRs, issues, tests, and dependencies, and serve cited context through CLI and MCP.

Acceptance outcome: A coding agent can request an evidence-backed context packet for a function.

Following

Phase 2 — Durable anchors and backlog

Resolve semantic anchors across commits, integrate GitHub Issues first, and explain task staleness when code or assumptions change.

Acceptance outcome: A task remains attached after a move or rename and is flagged when assumptions need revalidation.

Following

Phase 3 — Risk and regression memory

Add change coupling, churn, reverts, fixes, CI/test history, incidents, and evidence-scored regression relationships.

Acceptance outcome: A context packet explains historical risk and recommends targeted validation.

Following

Phase 4 — Multi-agent coordination

Add a local/team coordination service, semantic claims, overlap detection, contract warnings, refresh, handoff, and expiry.

Acceptance outcome: Agents receive warnings before semantic conflict becomes an integration failure.

Later

Phase 5 — Decision memory and scale

Import transcripts, reconstruct ADR drafts, broaden language coverage, and add enterprise deployment, policy, SSO, and cross-repository service maps.

Acceptance outcome: Teams use KodeAtlas as a durable software-memory layer across repositories and agent systems.

What waits

Protect the product from unproven breadth.

  • broad multi-language support before TypeScript/JavaScript lineage is validated;
  • automatic ADR publication without human review;
  • hard agent locks or mandatory centralized orchestration;
  • causal claims based only on temporal proximity;
  • a graph-database deployment before local query patterns justify it;
  • a polished enterprise control plane before context packets demonstrate value.

Take the next step

Prove the context packet before expanding the platform.

The immediate experiment is to index symbols in one archived TypeScript/JavaScript repository, connect them to existing evidence, and measure whether the context changes an agent’s plan.

Join the Phase 1 evaluation Review capability status