ChronosGraph — visual oversight and shared context

ChronosGraph

See what your coding agents change—and the evidence behind it.

Explore code relationships, understand project decisions, follow agent handoffs,
and review exact proposed changes in a FalkorDB-powered workspace.

Try it

• Watch the product walkthrough · 4:34
• Download the current source
• Run locally
• Read the documentation

The walkthrough shows actual product navigation through a prepared synthetic
project with recorded agent activity. The proposed change remains pending human
approval. Captions
and narration script
are included.

Also watch

Real-bug case study · 3:26:
a stale-context UI defect reproduced on the old build and rejected by the fixed
build using controlled browser regression inputs.

Version, verification and additional assets

• Current source: a7b923a489bd84d5b99d3eccf30d168fc9e288b9.
• Running application: 44bb68b459f83dbdc091ceecf4cd60ded72b379b.
• The original release tag remains pinned to 384e0e1; it has not been rewritten.
• Documentation-only updates do not change the running application.
• Scoped application checks: 47 frontend tests, typecheck, isolated production
  build, desktop/narrow browser walkthrough and context-invalidation regressions.
• Videos passed full decoding, checksum and anonymous browser playback checks.
• Supporting recordings, scripts, source manifests and earlier source archives
  remain attached with their original scope. The walkthrough is not a live
  OpenCode terminal/UI split-screen.

Current release scope ·
Limitations and evaluation ·
Credits and project notes.

ChronosGraph — Integrated Hackathon Release

ChronosGraph — integrated hackathon release

The primary main branch now contains the latest hackathon application, integration,
tests, benchmark reports and synchronized documentation from build/final-hackathon-release.
Verified application baseline: 479e987; publication/docs commit: dd8f44a.

Full developer walkthrough

Attached: 23-minute MP4 with 21 chapters and burned captions, PDF guide/complete
transcript, and SRT captions. Covers architecture, imports, source/dependency graphs,
MCP connection, broken/coordinated proposals, exact approval, observed trusted-local
pytest, source-bound memory, shortestPath replay, recovery, ownership, vault export,
policy invalidation, real repositories, benchmarks and intended daily use.

Real UI actions on an isolated fixture; scripted real MCP client; synthetic narration;
paced actions and held screens. This video is not a live autonomous-model session.
The separate real-model OpenCode verification remains in the repository reports.

Verification and limits

Latest full application batch: 333 backend tests (no failures/skips), 34 frontend unit
tests, typecheck/build, 118 browser check executions (96 excluding repeated recovery/setup).
Those receipts belong to baseline 479e987; this publication changes documentation only.
Benchmark acceptance remains NOT MET: large-repository lookup latency, legacy GraphRAG
import omissions and unmeasured runtime test-impact precision are documented, not hidden.
Focused byte reduction is not billed token savings or proof of safe test skipping.

External OpenCode/harness binaries and credentials are not included. Project-specific
setup, integration tests and evidence are included. Media is attached here rather than
committed into source history. Updating GitHub does not deploy the VM preview, publish
the blog externally or submit the project to the organizer.