{"slug": "agency-open-source-platform-for-debugging-profiling-and-analyzing-ai-agents", "title": "Agency: Open-source platform for debugging, profiling, and analyzing AI agents", "summary": "UC Berkeley Sky Computing Lab released Agency, an open-source platform for running, debugging, profiling and controlling AI agents, which the project describes as \"htop + gdb for AI agents.\" Agency links LLM requests, tool calls and system activity in one dashboard, adds a Critical Path Profiler to show where wall-clock time went, and compares completed runs side by side; it is model- and harness-agnostic, supporting Claude Code, Codex, OpenCode and a native harness across OpenAI, Anthropic, Amazon Bedrock and OpenAI-compatible endpoints including vLLM. The platform requires Linux x86-64, Python 3.12+, Git, uv, and Docker or Podman, with no GPU, ZFS or CRIU needed, and integration with existing Claude Code and Codex CLI sessions on Linux is labeled experimental.", "body_md": "Live debugging and systems profiling for AI agents.\n\nAgency is an open-source platform developed at [UC Berkeley Sky Computing Lab](https://sky.cs.berkeley.edu/) for **running, debugging, profiling, and controlling AI agents**.\n\nAs agents tackle longer tasks, invoke tools, and work in parallel, understanding what they're actually doing gets harder. Model traces only tell part of the story. What is an agent waiting for? Where is time being spent? Why is one run slower than another?\n\nAgency connects **LLM requests, tool calls, and system activity** in one place, giving you the visibility to investigate execution, find bottlenecks, and understand what changed.\n\nWe are model and harness agnostic! Run Claude Code, Codex, OpenCode, or Agency's native harness with supported backends including OpenAI, Anthropic, Bedrock, and local models via vLLM. [Get started](#getting-started)\n\nThink **htop + gdb for AI agents**.\n\nWatch agents execute in real time. Inspect model requests, tool calls, outputs, failures, and waits across agents. Pause, resume, or stop supported executions directly from the dashboard.\n\nWhen execution finishes, the same view becomes a navigable history of everything that happened.\n\nSee where wall-clock time actually goes.\n\nTrace model requests, tool calls, processes, and dependencies across agents. Identify the work determining completion time, distinguish parallel execution from waiting, and zoom into bottlenecks.\n\nUse the **Critical Path Profiler** after execution to investigate what held the run back.\n\nDig into the measurements behind execution.\n\nExplore LLM latency, token usage, tool durations, errors, retries, CPU, memory, I/O, and optional GPU metrics. Follow linked records from high-level activity down to individual events and processes.\n\nNo invented numbers. Missing measurements stay missing.\n\nCompare completed runs side by side.\n\nExamine execution trajectories, model and tool activity, latency, resource usage, failures, and outcomes. Find where execution paths diverged and which measurements changed.\n\nAgency compares recorded evidence, without guessing at causes.\n\nAgency provides a sandboxed execution environment, Python API, supported agent harnesses, and integrated telemetry. Your application keeps control of its workflow and orchestration.\n\n| Layer | Supported options | \n|---|---|\n| Agent harnesses | Native, Claude Code, Codex, OpenCode, Grok, Kimi | \n| Model backends | OpenAI Chat Completions / Responses, Anthropic, Amazon Bedrock, OpenAI-compatible endpoints (including vLLM) | \n| Environment | Linux sandboxes using Docker or Podman, with configurable resources and mounts | \n\nCompatibility depends on model capabilities, configuration, and external CLI versions. External harness CLIs must be installed separately.\n\n**Experimental:** Integration with existing Claude Code and Codex CLI sessions on Linux is experimental. Profiling, interception, and execution controls may be limited. Python-orchestrated workflows currently provide the most complete experience.\n\nSee [configuration and requirements](https://github.com/agency-project/agency/blob/master/docs/guides/configuration.md).\n\n- Linux x86-64 and Python 3.12+\n- Git and [uv](https://docs.astral.sh/uv/getting-started/installation/)\n- Docker or Podman\n- API key for your configured model provider\n\n```\ngit clone https://github.com/agency-project/agency.git\ncd agency\nuv sync --locked --python 3.12\n\nexport OPENAI_API_KEY=\"your-api-key\"\nexport OPENAI_MODEL=\"gpt-6-luna\"\n\nuv run python examples/quickstart.py\n```\n\nThis runs a native agent in an isolated sandbox. The configured model must support `reasoning_effort=\"none\"`. Model API usage may incur charges.\n\nFirst startup requires network access for the sandbox image and dependencies. No GPU, ZFS, or CRIU required.\n\n```\nAGENCY_PROFILE=1 AGENCY_PROFILE_SCOPE=workload \\\n  uv run python examples/quickstart.py --webui\n```\n\nOpen [localhost:7860](http://localhost:7860/) for live execution and [localhost:7860/profiler](http://localhost:7860/profiler) for profiling.\n\nThe dashboard binds to `0.0.0.0` by default; don't expose it to untrusted networks without appropriate access controls.\n\nFor Python integration, see [`examples/quickstart.py`](https://github.com/agency-project/agency/blob/master/examples/quickstart.py). You can configure the harness and model independently, define tasks with typed inputs and outputs, and collect execution records for later analysis.\n\n- [Getting Started](https://github.com/agency-project/agency/blob/master/docs/getting-started.md) — Installation and first run\n- [Configuration](https://github.com/agency-project/agency/blob/master/docs/guides/configuration.md) — Harnesses, models, and environments\n- [Profiling](https://github.com/agency-project/agency/blob/master/docs/guides/profiling.md) — Recording and interpreting execution\n- [Examples](https://github.com/agency-project/agency/blob/master/examples/README.md) — Agent workflows and integrations\n- [API Reference](https://github.com/agency-project/agency/blob/master/docs/api/index.md) — Python interfaces\n- [Architecture](https://github.com/agency-project/agency/blob/master/docs/architecture/index.md) — Runtime and system design\n\nAgency is open source. Contributions, bug reports, and feedback are welcome.\n\nSee [CONTRIBUTING.md](https://github.com/agency-project/agency/blob/master/CONTRIBUTING.md) to get started.", "url": "https://wpnews.pro/news/agency-open-source-platform-for-debugging-profiling-and-analyzing-ai-agents", "canonical_source": "https://github.com/agency-project/agency", "published_at": "2026-10-21 21:56:31+00:00", "updated_at": "2026-10-21 22:13:25.697309+00:00", "lang": "en", "topics": ["ai-agents", "ai-tools", "developer-tools", "mlops", "artificial-intelligence"], "entities": ["UC Berkeley Sky Computing Lab", "Agency", "Claude Code", "Codex", "OpenCode", "OpenAI", "Anthropic", "Amazon Bedrock"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/agency-open-source-platform-for-debugging-profiling-and-analyzing-ai-agents", "markdown": "https://wpnews.pro/news/agency-open-source-platform-for-debugging-profiling-and-analyzing-ai-agents.md", "text": "https://wpnews.pro/news/agency-open-source-platform-for-debugging-profiling-and-analyzing-ai-agents.txt", "jsonld": "https://wpnews.pro/news/agency-open-source-platform-for-debugging-profiling-and-analyzing-ai-agents.jsonld"}}