Agency: Open-source platform for debugging, profiling, and analyzing AI agents

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.

Live debugging and systems profiling for AI agents. Agency 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 . As 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? Agency connects LLM requests, tool calls, and system activity in one place, giving you the visibility to investigate execution, find bottlenecks, and understand what changed. We 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 Think htop + gdb for AI agents . Watch 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. When execution finishes, the same view becomes a navigable history of everything that happened. See where wall-clock time actually goes. Trace model requests, tool calls, processes, and dependencies across agents. Identify the work determining completion time, distinguish parallel execution from waiting, and zoom into bottlenecks. Use the Critical Path Profiler after execution to investigate what held the run back. Dig into the measurements behind execution. Explore 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. No invented numbers. Missing measurements stay missing. Compare completed runs side by side. Examine execution trajectories, model and tool activity, latency, resource usage, failures, and outcomes. Find where execution paths diverged and which measurements changed. Agency compares recorded evidence, without guessing at causes. Agency provides a sandboxed execution environment, Python API, supported agent harnesses, and integrated telemetry. Your application keeps control of its workflow and orchestration. | Layer | Supported options | |---|---| | Agent harnesses | Native, Claude Code, Codex, OpenCode, Grok, Kimi | | Model backends | OpenAI Chat Completions / Responses, Anthropic, Amazon Bedrock, OpenAI-compatible endpoints including vLLM | | Environment | Linux sandboxes using Docker or Podman, with configurable resources and mounts | Compatibility depends on model capabilities, configuration, and external CLI versions. External harness CLIs must be installed separately. 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. See configuration and requirements https://github.com/agency-project/agency/blob/master/docs/guides/configuration.md . - Linux x86-64 and Python 3.12+ - Git and uv https://docs.astral.sh/uv/getting-started/installation/ - Docker or Podman - API key for your configured model provider git clone https://github.com/agency-project/agency.git cd agency uv sync --locked --python 3.12 export OPENAI API KEY="your-api-key" export OPENAI MODEL="gpt-6-luna" uv run python examples/quickstart.py This runs a native agent in an isolated sandbox. The configured model must support reasoning effort="none" . Model API usage may incur charges. First startup requires network access for the sandbox image and dependencies. No GPU, ZFS, or CRIU required. AGENCY PROFILE=1 AGENCY PROFILE SCOPE=workload \ uv run python examples/quickstart.py --webui Open localhost:7860 http://localhost:7860/ for live execution and localhost:7860/profiler http://localhost:7860/profiler for profiling. The dashboard binds to 0.0.0.0 by default; don't expose it to untrusted networks without appropriate access controls. For 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. - Getting Started https://github.com/agency-project/agency/blob/master/docs/getting-started.md — Installation and first run - Configuration https://github.com/agency-project/agency/blob/master/docs/guides/configuration.md — Harnesses, models, and environments - Profiling https://github.com/agency-project/agency/blob/master/docs/guides/profiling.md — Recording and interpreting execution - Examples https://github.com/agency-project/agency/blob/master/examples/README.md — Agent workflows and integrations - API Reference https://github.com/agency-project/agency/blob/master/docs/api/index.md — Python interfaces - Architecture https://github.com/agency-project/agency/blob/master/docs/architecture/index.md — Runtime and system design Agency is open source. Contributions, bug reports, and feedback are welcome. See CONTRIBUTING.md https://github.com/agency-project/agency/blob/master/CONTRIBUTING.md to get started.