Skip to main content
LoopGuard watches an AI agent while it runs and checks whether it is still making progress. It detects loops, repeated failures, and stalled runs while allowing useful polling and retries to continue.
These are patterns to look for, not guarantees about a fixed number of calls. Decisions depend on the active detectors, thresholds, and recorded evidence.

Get Started

Install the SDK and catch a stuck tool loop without an API key.

View GitHub

Explore the Python SDK, examples, and tests.

​
See the problem

LoopGuard demo with a red agent-tool loop, semantic repetition evidence, and an interrupted-run verdict

Screenshot from a LoopGuard demo run showing repeated intent and an interrupted run.

The visual examples and SDK code show how to inspect the same detection behavior from Python.

​
Where LoopGuard fits

LoopGuard focuses on whether an agent is still making progress. It is not an agent framework, a general observability platform, a tracing replacement, or an LLM evaluation platform. Keep your existing agent and tracing tools; add LoopGuard where you need evidence about repetition and stuck runs. The Python package includes LangGraph/LangChain tool callbacks, a live-stream wrapper, and saved-trace analysis. The local demo helps you compare healthy repeats with runs that should stop.
The published 0.2.1 package supports the core callback and trace workflows. Gemini support, controlled demo scenarios, and additional decision display fields currently require the source checkout. See release scope.

Integrate with your agent

Attach callbacks and handle runtime stop decisions.

Understand the pipeline

Follow an event from the agent to a runtime decision.

Recognize stuck behavior

Compare repetition, failure, no progress, and cycles.

Compare demo scenarios

See healthy recovery beside runs that should stop.
Already running into a problem? Start with troubleshooting. For background, read AI agents and loops, tests, traces, and evals, and agent trajectories and convergence.

​
Ask about LoopGuard

Use the menu beside the page title to open the current documentation page in ChatGPT, Claude, Google AI Studio (Gemini), Perplexity, or Grok. These links open an external AI tool using your own account and include the page as context for questions. Try asking how to attach the callback, why a repeated tool call didn’t stop, or which features are available in the published package. Start from the relevant guide or reference page and ask for links to the supporting documentation. For a project-wide question, include the published site’s llms.txt index or llms-full.txt export. If the tool cannot fetch a page, use Copy page from the same menu and paste the Markdown into your conversation. Public page fetching needs the deployed docs URL; an external tool cannot read your localhost preview.