Back to directory

Agent Flow

A real-time visualizer that turns Claude Code and Codex agent activity into an interactive node graph, so a developer can watch how an agent breaks down and executes a task.

Agent Flow

Agent Flow shows Claude Code and Codex agent orchestration as it happens, drawn as an interactive node graph the developer can pan, zoom and click into. It auto-detects sessions from both agents concurrently, or can be restricted to one, and shows how an agent breaks a problem down, which tools it calls, and how any subagents branch and coordinate.

Claude Code events stream in through a lightweight local HTTP hook server for close to zero latency, while Codex activity is read by tailing its rollout log files, surfacing tool calls, reasoning and token counts from Codex's own event stream; the tool also accepts any JSONL event log to replay. Multiple concurrent sessions appear as separate tabs, and timeline, file-attention and transcript panels let a developer trace exactly which tool calls were slow or redundant. It runs as a standalone web app launched with a single command, or as a VS Code extension, and was built while its author was developing a separate AI-driven game creation platform.

More in AI evals

See all 269
  • Provenance gate

    github.com

    A deterministic authorization gateway that sits between an AI agent and its tools, tracing where each parameter of a side-effect call came from, and blocking the call or asking a human when the source does not check out.

  • JevBench

    benchmarkheaven.com

    A reproducible benchmark from Benchmark Heaven that ranks Jev-class typed decision models on intelligence, calibration, speed, and cost, with a published harness, scoring rules, and results.

  • A co-evolutionary adversarial red-teaming tool for AI agents that uses quality-diversity evolutionary search to autonomously breed edge-case scenarios, testing whether an agent with tool access can be made to hallucinate, deadlock or leak funds.