Analysis updated 2026-05-18
Run an AI coding agent in read-only mode to get an architecture summary of a codebase safely.
Delegate a task to supervised subagents and inspect exactly what each one did afterward.
Resume a previous agent session and continue where it left off.
Audit a failed or partially completed agent task through replayable local logs instead of trusting a summary.
| ranvier-technologies/pixir | caudena/beam_weaver | carterperez-dev/exs-cyberjob-scraper | |
|---|---|---|---|
| Stars | 20 | 20 | 21 |
| Language | Elixir | Elixir | Elixir |
| Setup difficulty | moderate | moderate | easy |
| Complexity | 4/5 | 4/5 | 2/5 |
| Audience | developer | developer | general |
Figures from each repo's GitHub metadata at analysis time.
Requires Elixir 1.20 or newer and either a ChatGPT subscription login or an OpenAI API key.
Pixir is a runtime that sits underneath AI coding agents and keeps a careful, auditable record of what they did, even when the work is long, delegated to smaller sub tasks, or fails partway through. Instead of treating the chat window as the source of truth, Pixir treats itself as the actual runtime: the chat interface or tool you use is just a presenter asking Pixir to do work, while Pixir is the part that actually runs the tools, supervises the process, and keeps evidence of what happened. This distinction matters once an agent's work becomes more complex than a single quick reply. If an agent hands off part of a task to a smaller sub agent, that sub agent needs its own tracked lifecycle. If a multi step task only partly succeeds, that partial result needs to be recorded honestly rather than hidden. Pixir keeps local, replayable logs of every session, so a person or another agent can later inspect exactly what ran, what tools were used, and what failed or timed out, rather than trusting a written summary after the fact. You can run Pixir directly from the command line inside any project, starting in a read-only mode that only inspects and explains a codebase without making changes, or in a mode that asks permission before writing files or running commands that could be unsafe. It can resume a previous session, and it has commands to inspect a session's replay log, diagnose problems, or view a tree of subagent activity. It also connects to editors and tools that speak the Agent Client Protocol, acting as the runtime behind those interfaces. Pixir authenticates either through a ChatGPT subscription login or an OpenAI API key. It is described by its own authors as a developer preview, meaning its command line and protocol behavior are the intended stable surface, but its internal library code is not yet promised to stay stable. This is aimed at developers and power users running AI coding agents who want a dependable, inspectable record of what those agents actually did, especially when work is delegated across multiple steps or subagents. It is built in Elixir and installs through the Hex package manager or from source, requiring Elixir version 1.20 or newer.
A runtime that supervises AI coding agent work, keeping an inspectable, replayable log of sessions, subagents, and partial or failed outcomes.
Mainly Elixir. The stack also includes Elixir, OTP, Phoenix.
No license information was found in the README excerpt.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.