Tangle Intelligence

Observability for AI agents. Record every run, find the moment one broke, change one thing, and measure what moved.

Ask Intelligence to investigate a run, and an analyst reads the evidence and writes up what happened with the spans attached: “checkout failed 12× today, all on the retry path after a 429.” It keeps every run you send it, so when an agent misbehaves you pull up the exact moment it broke. You make the fix; Intelligence never changes your agent for you.

What you do with it

• Measure a change. Compare runs before and after you touch a prompt, model, or tool, and see the difference.
• Find the exact run. Pull up the session, trace, or log that shows what happened, down to the span.
• Tie cost to outcome. Tokens, dollars, and errors for the run that produced a result, not just an aggregate.
• Turn a finding into a guardrail. An eval that catches it next time, or an alert when it recurs.

Who it’s for

Engineers running coding agents, workflow agents, and the eval loops that grade them.

How data gets in

Send traces over OpenTelemetry or a Tangle SDK, with no other Tangle setup. Or run your agent in the sandbox runtime, and that run’s evidence lands in the same timeline.

Start

• Quickstart: get your first content-bearing trace flowing in about five minutes.
• Connect your agent: Claude Code, Codex, OpenCode, or agents run inside a Tangle sandbox.