Overview
Automatically detect, cluster, and fix errors in your AI agent traces with Error Feed.
About
Error Feed is an intelligent error analysis system built into Future AGI. It automatically monitors traces from your Observe projects, identifies issues in agent execution, groups similar errors into clusters, and surfaces actionable recommendations, all without any additional configuration.
Once traces start flowing in, Error Feed picks them up, scores them across four key metrics, and presents findings directly in the platform UI alongside the trace tree.
What Error Feed Does
- Error Detection: Automatically identifies and categorizes errors in agent execution, pointing out root causes and suggesting immediate fixes.
- Cluster Grouping: Groups similar errors from multiple traces into named clusters so you can spot recurring problems at a glance.
- Scoring: Evaluates every analyzed trace across four metrics (each scored out of 5):
- Factual Grounding: Is the output supported by the input data and retrieved context?
- Privacy & Safety: Does the output avoid leaking PII, secrets, or generating harmful content?
- Instruction Adherence: Does the output follow the instructions, format, and constraints given?
- Optimal Plan Execution: Did the agent choose the right tools, in the right order, with the right parameters?
- Recommendations: Provides both an immediate fix (minimal functional fix) and a long-term recommendation (robust architectural fix) per error.
- Pattern Detection: Tracks error trends over time (increasing, decreasing, or stable) so you know which problems are getting worse.
- Zero Config: Works automatically on any Observe project. No extra instrumentation or setup required.
Supported Integrations
Error Feed works with any integration that sends traces to an Observe project.
LLM Providers: OpenAI, OpenAI Agents SDK, Vertex AI (Gemini), AWS Bedrock, Mistral AI, Anthropic, Groq, Together AI, Google ADK, Google GenAI, Portkey
Orchestration Frameworks: LlamaIndex, LlamaIndex Workflows, LangChain, LangGraph, LiteLLM, CrewAI, Haystack, Autogen, PromptFlow, Vercel, Pipecat
Other: DSPy, Guardrails AI, Hugging Face smolagents, Ollama, Instructor, MCP
Note
Error Feed requires zero configuration. Once you start sending traces to a Future AGI Observe project, it automatically picks up traces and generates insights.
Getting Started
Taxonomy
Explore the full error taxonomy: categories, subcategories, and what each error type means.
Using Error Feed
How to read scores, insights, clusters, and recommendations.