What is Mibo?
Mibo lets you test what your agent says and does. Your Agent sends Mibo a trace of each execution. Mibo checks the recorded responses and tool calls against your rules and shows the evidence behind each result.
Mibo never calls your Agent. It evaluates only the traces you send.
The problem Mibo solves
Section titled “The problem Mibo solves”AI systems are unpredictable. The same question can produce different answers every time. They might call the wrong tools, make up information, or give unhelpful responses.
Manual review doesn’t scale. You can’t read every conversation after each update. When something breaks, you usually find out from frustrated users, not from a dashboard.
Mibo checks the traces you send against your tests, so you can see where your Agent breaks your rules.
How Mibo works
Section titled “How Mibo works”- Send your first trace. Configure an OTLP exporter, the n8n community node, or a direct HTTP request, and capture an execution from your system.
- Generate a test with AI. Describe a scenario, or select a completed trace, and ask the Test Architect to draft the checks. Review the draft before saving it.
- Evaluate future traces. Send later executions. Mibo evaluates each new trace against your enabled tests and shows which checks failed, together with the recorded response or action.
Results show which checks passed and which failed. Each failure gives its reason and the recorded response or action behind it.
Key concepts
Section titled “Key concepts”Workspace: the container for your Agents, tests, API keys and results. Mibo creates it when you sign up.
Agent: the AI system whose traces Mibo evaluates. One workspace can have several Agents, for example staging and production.
Trace: the record of one execution of your Agent. It contains spans for the steps your Agent took, such as the model response and each tool call. See the Canonical trace reference.
Test case: a scenario you want to verify. For example: “If a user asks about store hours, the agent responds with the correct schedule.” Mibo evaluates each new trace against the test cases that apply to it.
Assertion: a single check within a test case. Assertions come in two kinds. Procedural checks are rule-based: did the Agent call the right tool, or avoid a forbidden node? Semantic checks are scored by an AI evaluator: was the response helpful, on topic, and free of hallucination?
Run: one evaluation of a trace against the tests that apply to it.
Who is Mibo for?
Section titled “Who is Mibo for?”- Teams building AI agents who want to ship with confidence.
- Product managers who need visibility into AI quality before and after releasing updates.
- QA teams who want to check real conversations instead of clicking through scenarios manually.
The rest of these docs follow the order you’ll use Mibo: set up an Agent, send traces, create tests, and review the results.