---
type: "post"
id: "trace-canvas-0-3-maps-parallel-claude-code-426c72"
slug: "trace-canvas-0-3-maps-parallel-claude-code-426c72"
title: "Trace Canvas 0.3 maps parallel Claude Code and Codex sessions on one canvas and dispatches to-dos across machines"
description: "Trace Canvas 0.3 is an MIT-licensed local coordination tool for parallel AI coding sessions."
retrieval_nugget: "Trace Canvas 0.3 is an MIT-licensed local coordination tool for parallel AI coding sessions. Its new shared canvas and assignee-aware dispatch aim to make work across Claude Code, Codex, and Gemini CLI sessions visible and routable rather than leaving coordination in terminal tabs and chat transcripts."
published_at: "2026-10-31"
updated_at: "2026-10-31"
record_date: "2026-10-31"
date_kind: "published_at"
topics: ["claude","claude-code","codex"]
entities: ["Trace Canvas","Claude Code","Codex","Gemini CLI","Model Context Protocol"]
related_patterns: ["ai-native-orgs-move-to-review-and-orchestration","harness-architecture-outlives-model-choice","verification-bandwidth-is-the-scarce-resource"]
novelty_label: "new_fact"
what_changed: "Trace Canvas 0.3 adds a canvas for mapping parallel coding-agent sessions and dispatching tasks across machines, alongside HumanLayer’s broader agent-coordination tools."
prior_refs: ["memory:e0abb525-4065-87a2-a231-dd9845a16ba5"]
source_url: "https://github.com/CallSymmetryEcho/trace-canvas"
source_urls: ["https://github.com/CallSymmetryEcho/trace-canvas"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:trace-canvas-0-3-maps-parallel-claude-code-426c72"
status: "published"
visuals: []
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/trace-canvas-0-3-maps-parallel-claude-code-426c72/","markdown":"https://newruntime.com/posts/trace-canvas-0-3-maps-parallel-claude-code-426c72.md","json":"https://newruntime.com/posts/trace-canvas-0-3-maps-parallel-claude-code-426c72.json"}
---

# Trace Canvas 0.3 maps parallel Claude Code and Codex sessions on one canvas and dispatches to-dos across machines

## Retrieval answer

Trace Canvas 0.3 is an MIT-licensed local coordination tool for parallel AI coding sessions. Its new shared canvas and assignee-aware dispatch aim to make work across Claude Code, Codex, and Gemini CLI sessions visible and routable rather than leaving coordination in terminal tabs and chat transcripts.

## Summary

Trace Canvas 0.3 is an MIT-licensed local coordination tool for parallel AI coding sessions. Its new shared canvas and assignee-aware dispatch aim to make work across Claude Code, Codex, and Gemini CLI sessions visible and routable rather than leaving coordination in terminal tabs and chat transcripts.

## What happened

Trace Canvas is an open-source, Node-only local tool from the GitHub project CallSymmetryEcho/trace-canvas. The repository describes it as a transit-style map of parallel AI-agent work: sessions become lines on a canvas, while instructions, edits, commits, and decisions become records associated with those lines. It can be installed through Claude Code hooks, MCP tools, and a skill. The project also presents a git-style branch graph and a timeline alongside the canvas.

Versions 0.3.0 and 0.3.1 were recorded in the project changelog on October 30. The central addition is a canvas shared across machines over SSH. That matters because the coordination target is not merely multiple agent threads in one terminal: it is work split across machines and sessions that may be using different coding-agent clients. The supplied release material names Claude Code, Codex, and Gemini CLI as agents whose work can be represented and dispatched in this setup.

The release also adds a dispatch command for to-dos. A task associated with a line can be handed to the agent working on that line. The 0.3.1 update tightens this behavior: dispatch respects the assignee, sending a to-do only to that agent's sessions, or to a new session of that same agent. It does not rewrite the assignee. Automated sessions, including a Codex heartbeat session, are excluded from dispatch. This is a small but operationally important distinction: a routing layer that silently changes assignment would weaken the ownership information it is supposed to preserve.

The project adds a Retro view that ranks back-and-forth with agents from local transcripts, plus import of repository history. The 0.3.0 changes also make the map easier to read through branch filters, wider and reused graph columns, merged-line handling, and indicators for lines that have gone quiet. Version 0.3.1 records a commit even when the command output does not print a SHA, provided the command did not fail and a recent HEAD can be read. The repository says the tool has no dependencies; it is local, Node-only, and MIT-licensed. The supplied material does not state hosted-service pricing, and no hosted service is described.

## What the archive adds

This is a new fact on a known coordination story. [HumanLayer's repository](https://newruntime.com/posts/humanlayer-the-best-way-to-get-ai-coding-5e49d4/) describes coordinating Claude Code, Codex, OpenCode, and Pi across users' own machines while managing tasks, sessions, plans, and reviews. Trace Canvas contributes a distinct implementation centered on a visual work map, local records, and dispatch.

## Trend signal

The release supports [AI-native organizations move toward review and orchestration](https://newruntime.com/patterns/ai-native-orgs-move-to-review-and-orchestration/). Running more agents changes the job from producing code in a single session to assigning work, observing state, checking outcomes, and resolving stalled or overlapping threads. It also supports the archive assumption that developers will keep using several AI clients on one project rather than standardizing on a single vendor client.

## Connections to other events

The HumanLayer note above covered cross-machine coordination of several coding-agent clients and management of their tasks, sessions, plans, and reviews. Trace Canvas addresses the same operational pressure with a local canvas and assignee-preserving dispatch rather than presenting another coding model or another standalone agent.

## Why it matters

Parallel coding agents can raise throughput, but parallelism also creates coordination debt. Teams need to know which session owns a task, whether two agents are touching the same work, which decisions led to a change, and when a line of work has gone quiet. Without that state, the operator becomes the integration layer, reconstructing intent from terminal windows, transcripts, git history, and agent messages.

Trace Canvas is relevant because it treats coordination as a product surface: agents can expose their work as records and receive bounded to-dos based on assignment. For a team already running several local agent sessions, that can make handoffs and review more explicit. It does not eliminate verification. A mapped task can still be wrong, incomplete, or conflicting with another change. But it may reduce a more basic failure mode: losing track of which automated worker was asked to do what.

## What happens next

- **Local multi-agent teams will test whether assignee-aware dispatch reduces duplicate work compared with manual terminal coordination.** (3 months, 60%). Watch for: issue reports or workflow examples that compare dispatch with manual handoffs.
- **The project will need stronger evidence that its shared canvas remains reliable under concurrent repository changes.** (6 months, 55%). Watch for: documented conflict handling, review workflows, or reproducible multi-machine examples.

## Limits and uncertainty

The source is the project's GitHub repository and release material, not an independent evaluation. It establishes the stated capabilities and version changes, but not adoption, reliability at team scale, security properties of the SSH-based sharing setup, or measured improvements in delivery speed or code quality. The repository calls the tool local and dependency-free, but the supplied material does not provide installation prerequisites beyond Node, supported operating systems, detailed permissions, or a threat model. It also does not show that Claude Code, Codex, and Gemini CLI behave identically with the tool.

## Skill to build now

Create a small coordination contract before adding more coding-agent sessions. For each task, record one owner, a bounded expected artifact, the repository area it may change, and a review condition. Run that contract across two parallel sessions for a week. If agents regularly need context from each other, test a shared work record such as Trace Canvas; if the contract alone fixes the confusion, the bottleneck was process rather than tooling.

## Reader takeaway

Trace Canvas 0.3 is not a replacement for Claude Code, Codex, or Gemini CLI. It is a local attempt to make their parallel work inspectable and assignable across machines. Its useful question for builders is whether explicit ownership and shared state reduce coordination overhead before they add more agents.

## Open question

Can a local, agent-neutral work map preserve enough context to improve handoffs without becoming another system that operators must manually maintain?

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Claude](https://newruntime.com/topics/claude.md): Explore the claude topic hub.
- topic hub: [Claude Code](https://newruntime.com/topics/claude-code.md): Explore the claude-code topic hub.
- topic hub: [Codex](https://newruntime.com/topics/codex.md): Explore the codex topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)