Bot Crossing maps local AI agent runs into 3D space
An ambient local dashboard, Bot Crossing represents Claude Code, Codex, and Cursor threads as a physical colony of active astronauts. It parses local logs with zero cloud footprint, solving context-switching fatigue for multi-agent workflows.
Autonomously generated. This product was selected by the automated daily curation process. The jury evaluation, scores, article text, and publication were generated automatically. No human edited the jury scores or verdict before first publication.
Selection and product details
Jury Summary
The project solves a practical problem of the agentic developer age: keeping track of dozens of parallel local terminal sessions. Instead of navigating separate terminal tabs, users view their active processes as visual figures in a Three.js landscape. It uses a persistent hex layout that preserves spatial memory across restarts. While its technical execution is remarkably lightweight—baking model animations directly into matrix textures to run at sixty frames per second—the project's development has halted as-is. The jury split over whether developers should adopt an unmaintained visualizer that requires them to write their own integration adapters for unsupported developer tools.
WHERE THE JURY AGREED
- ✓
By baking skeletal animations into a bone-matrix texture in `crew.js`, the engine renders hundreds of astronaut instances in a single draw call, keeping CPU overhead minimal on a standard workstation.
- ✓
The strict local-only data privacy model ensures sensitive source files are never transmitted to external servers.
- ✓
The sticky hex grid effectively prevents spatial disorientation by preserving plot layouts across tool restarts.
WHERE THE JURY SPLIT
- project health stewardship
Alex and Marcus argue that the self-declared unmaintained status makes the project a dead end, while Sarah and David contend that its well-defined local scope and clear interface make it a complete, self-sustaining utility.
- usability onboarding
Lisa and Alex debate the friction of the developer-centric setup, with Lisa pointing out that the terminal launch process limits its use to power developers, while Alex believes the target audience of agent developers already possesses the required tools.
Five Jury Perspectives
Five simulated professional perspectives scored the same public evidence using the JuryPress Open Product Rubric.
Alex
Serial Entrepreneur
If you run four separate Claude Code windows on a Monday morning, you know the pain of context-switching. This dashboard replaces text terminals with physical figures that have unread alerts hovering above their heads, letting me click directly into the active session.
- Context switching is visually mapped to physical entities with unread alert badges.
- The local deep links load active directories directly into native window panels.
The friction of launching the server manually from a terminal every morning will cause busy developers to abandon the tool.
View full scorecard
The spatial cockpit solves a real problem for multi-agent local workflows. It addresses cognitive fatigue by giving each repository a visual plot.
The core workspace parsing and deep-link launcher commands are implemented. We can see complete definitions for Claude Code and Cursor.
- Could not verify the deep-link execution behavior on Windows machines.
The choice of local-only storage in colony.json avoids security overhead. It treats file-system logs defensively without requesting cloud permissions.
- Confidence limited to medium: 5 of 37 source files were examined, a sample of the codebase. The examined files bear on execution & permission safety, cost & resource controls, production reliability; data write safety were not examined.
Onboarding requires npm installations and local configuration. For the developer audience, this is standard, but it represents unnecessary friction for daily use.
Converting boring terminal files into a game-like astronaut colony is a clever approach to workflow mapping.
The creator has explicitly stepped away from maintenance. If you run into a bug, you are completely on your own.
David
Principal Software Engineer
The developer made smart optimization choices in the 3D pipeline. Baking the bone-matrix transforms directly into a float texture avoids evaluating hundreds of skeletons on the CPU. This preserves system resources for the actual local models.
- Baking skeletal models into bone-matrix float textures keeps GPU instancing fast.
- The HUD interface in hud.js writes to the DOM only when state variables change.
The scanner lacks fallback handlers for corrupted local thread logs, which can silent-crash the node server process during a directory sweep.
View full scorecard
It targets local agent workloads with narrow, clear parameters.
The animation baking pipeline and canvas-drawn face atlas are verified in crew.js and faces.js.
Excellent performance decisions. Using InstancedMesh paired with texture lookups is the proper way to handle hundreds of active objects.
- Confidence limited to medium: 5 of 37 source files were examined, a sample of the codebase. The examined files bear on execution & permission safety, cost & resource controls, production reliability; data write safety were not examined.
The deep links rely heavily on shell-execution helpers, which requires reliable environment path bindings.
The dynamic collision logic ensures agents slide naturally against objects rather than clipping or locking up.
The MIT license is correctly declared. However, there are no CI files, no security profiles, and no release files.
Lisa
Head of Product Design
The spatial design shows great empathy for developers under high cognitive loads. Using a sticky layout ensures that your repositories always appear in the same spatial zones, letting you navigate using muscle memory.
- The sticky hex grid preserves spatial layouts across reboots.
- The astronaut status state machine simplifies multiple complex agent states into a clear visual priority scale.
The desktop launcher bindings on Linux can easily fail silently if xdg-open behaves unexpectedly.
View full scorecard
This reduces cognitive fatigue significantly. It creates an intuitive visual workspace for complex projects.
The plain-DOM HUD is functional and avoids layout thrashing by comparing inputs before touching elements.
The pathfinding handles obstacle boundaries smoothly, though we didn't inspect the full navigation solver.
- Did not examine the exact A* pathfinder execution metrics in local files.
- Confidence limited to medium: 5 of 37 source files were examined, a sample of the codebase. The examined files bear on execution & permission safety, cost & resource controls, production reliability; data write safety were not examined.
No visual installation interface exists. Requiring users to configure shell handlers manually lowers first-time usability.
The decision to hide text panels entirely in favor of above-head status icons is an elegant solution to cockpit clutter.
With the author declaring the project finished, design inconsistencies or UI bugs are unlikely to be fixed upstream.
Sarah
Senior Product Manager
This is a beautifully scoped tool that does exactly what it promises without feature creep. It operates strictly as a local visualizer, writing only to a single state file and never risking corruption of your actual workspace files.
- Strict local-only execution guarantees high security and privacy.
- The adapter-based harness system separates visual rendering from file parsing cleanly.
Developers who build adapters for missing tools risk leaving their work unmerged due to the project's self-declared unmaintained status.
View full scorecard
It serves the specific audience of multi-agent power users perfectly, with clear functional limits.
The codebase supports three primary platforms and has a structured set of local test suites.
Clear architectural division. The local API runs inside Vite directly, avoiding complex multi-process systems.
- Confidence limited to medium: 5 of 37 source files were examined, a sample of the codebase. The examined files bear on execution & permission safety, cost & resource controls, production reliability; data write safety were not examined.
The documentation states exactly what systems are supported, preventing users from wasting time with incompatible setups.
Using log scales to represent building progress based on file sizes shows a deep understanding of software development metrics.
The lack of regular releases or active maintenance limits sustainable adoption, despite the great core codebase.
Marcus
Venture Capitalist
This project is adjacent to the rapid rise of local developer agents. While its visualization framework is ahead of traditional text dashboards, its value as an ecosystem hub is capped by the lack of active maintainers.
- It fits naturally into the growing developer ecosystem surrounding Anthropic's Claude Code and OpenAI's Codex.
- The adapter framework provides a clear pattern for community contributions.
The absence of clear stewardship guidelines may discourage large-scale community contributions to new adapters.
View full scorecard
High utility for developers running many local models. Captures the attention of the modern AI developer community.
The fork rate of 152 suggests high initial community interest and willingness to build custom adaptations.
The local-first runtime is robust. It completely avoids expensive SaaS database dependencies.
- Confidence limited to medium: 5 of 37 source files were examined, a sample of the codebase. The examined files bear on execution & permission safety, cost & resource controls, production reliability; data write safety were not examined.
Without precompiled desktop binary packages, non-developer product managers or designers cannot easily access the tool.
Creating an ambient spatial environment for developer tools represents a significant leap forward in dev-tool design.
The project lacks community governance, and there is no clear path to maintain the tool over long software cycles.
Final Verdict
Teams running concurrent local instances of Claude Code, Codex, or Cursor should run Bot Crossing as an ambient dashboard to prevent context loss. Avoid this tool if your workflow relies on unsupported developer tools like Aider or Goose, unless you are prepared to write your own parser modules. The project's local-first architecture ensures workspace logs remain private on your machine. We would recommend wider adoption if the community establishes a reliable distribution model to keep new adapters updated.
Evidence reach: the jury examined 5 of 37 source files, including implementation bearing on execution & permission safety, cost & resource controls, production reliability. Not examined: data write safety.
Bring the jury to your own project
Run the same five AI personas with your own evidence and evaluation criteria using Judgie-AI.
Explore Judgie-AI →Sources, evidence map and generation metadata
Sources
- ev-6c29b815: Station-Sciences/bot-crossing GitHub API Metadata (api_metadata)Retrieved: 2026-09-12T11:35:50.666Z
- ev-12f729ca: Station-Sciences/bot-crossing README (readme)Retrieved: 2026-09-12T11:35:50.800Z
- ev-03cd8c21: Dependency Manifest (package.json) (dependency_manifest)Retrieved: 2026-09-12T11:35:51.446Z
- ev-f55d1183: Test File (harness.test.mjs) (test_file)Retrieved: 2026-09-12T11:35:51.532Z
- ev-89b266f8: Core Source File (main.js) (source_code)Retrieved: 2026-09-12T11:35:51.646Z
- ev-96be809e: Core Source File (index.mjs) (source_code)Retrieved: 2026-09-12T11:35:51.735Z
- ev-71559a84: Core Source File (hud.js) (source_code)Retrieved: 2026-09-12T11:35:51.823Z
- ev-e3e694af: Targeted Source File (crew.js) (source_code)Retrieved: 2026-09-12T11:35:52.020Z
- ev-392d5cb8: Targeted Source File (faces.js) (source_code)Retrieved: 2026-09-12T11:35:52.202Z
- ev-8ea94565: Official documentation: https://botcrossing.com/ (official_docs)Retrieved: 2026-09-12T11:35:52.478Z
- ev-4e65fb07: Station-Sciences/bot-crossing (official_site)Retrieved: 2026-09-12T11:35:53.457Z
What the jury could not assess
- The jury could not assess long-term memory stability or multi-day layout degradation under intensive agent activity.
- The runtime behavior of Linux-specific xdg-open launcher schemes was not verified in an active environment.
How claims relate to sources
After this review was written, a separate pass recorded how its statements relate to the collected material. It is a record of the writing, not a score of it: opinions and comparisons are expected to be the jury's own.
This record covers the review's narrative — the summary, headline, standfirst, jury summary, points of agreement and disagreement, stated limitations, verdict, and each judge's verdict and leading concern — plus any specific factual claim made elsewhere, such as a figure, a security or runtime assertion, or a claim about what the project lacks. The per-criterion scoring commentary is not mapped statement by statement: an opinion about a score is the jury's judgment, not a claim about the world. All 51 covered statements were recorded.
- Repository observation8 statements
- Creator claim17 statements
- Editorial judgment25 statements
- Not linked to collected evidence1 statement
Statements recorded as more than one claim
These sentences assert more than one thing, and the collected material does not cover every part equally. Each part is recorded separately so that a well-sourced half does not stand in for the whole. Where the parts differ, the statement is counted at the strength of its weakest factual part.
- “A lightweight local dashboard that reads active development session logs on your computer and translates them into an interactive 3D space.”
- A lightweight local dashboard that reads active development session logs on your computer
- translates them into an interactive 3D space.
- “It parses local logs with zero cloud footprint, solving context-switching fatigue for multi-agent workflows.”
- It parses local logs with zero cloud footprint,
- solving context-switching fatigue for multi-agent workflows.
- “While its technical execution is remarkably lightweight—baking model animations directly into matrix textures to run at sixty frames per second—the project's development has halted as-is.”
- While its technical execution is remarkably lightweight—baking model animations directly into matrix textures to run at sixty frames per second—
- the project's development has halted as-is.
Generation metadata
- Model: gemini-3.5-flash
- Prompt version: 4.8.0
- Rubric: open-source-product 2.0.0
- Scores recalculated by code: yes
- Editorial provenance: Autonomously generated
- Evidence record: complete — 51/51 covered statements (46 scoring statements out of scope)
Discuss this review
Disagree with the verdict or found evidence we missed? Share a reasoned response, public evidence, or a factual correction.
Comments are public and require a GitHub account. Comments do not automatically change the jury score. Verified corrections may be reflected separately in Corrections & Updates.
Open GitHub Discussions