AI observability walkthrough
On this page
AI observability walkthrough
Runs, agents, MCP servers, models and cost for BoxLang AI. BoxLang+.
What it shows
- Overview: Runs, success, run time, tokens and cost.
- Needs attention: Problems found against your own limits.
- Agents: A dashboard for all of your agents.
- One card per agent: Model, tools, success rate, speed and issues.
- Open an agent: composition: How the agent is put together.
- Tools and parts: Calls, time and failures of every tool.
- Middleware: The stack, in the order it runs.
- Instructions, and the runs of the agent: A version for the prompt, and a filtered list of runs.
- Runs: Every run, of every kind.
- Open a run: the trace: Every step, nested the way it ran.
- MCP servers: The servers you host with BoxLang AI.
- Open an MCP server: Failures, tools and requests.
- Models: Calls, latency, tokens and errors per model.
- Cost, and where prices come from: An estimate, from prices you enter per million tokens.
- Registries: What is registered: agents, tools, skills and more.
- Settings: The recorder, the content and the memory.
- Alerts and prices: Your limits for needs attention, and the price table.
- What Free keeps: Three counters, last hour, counted only.
Read more
See all the walkthroughs.