Appstrate
[Infrastructure]

See every run, every cost, every error.

Full run traces. Cost per run. Tool call history. Structured logs. SSE streaming to your UI.

[01 · Why it matters]

You can't ship agents you can't see.

When an agent misbehaves, you need the trace: what did it decide, what tool did it call, what did the LLM return, how much did it cost.

Appstrate records every step: prompt, reasoning, tool calls, tool outputs, final output, cost. Queryable via API, streamable via SSE, exportable on Enterprise.


[02 · How it works]

Structured logs. Real-time. Full trace.

Every run records a timeline. Every container streams stdout JSON lines. Every LLM call logs its cost.

In practice

Every piece of Appstrate is a declared, versioned artifact — the agent, its tools, its skills, its provider connections. You describe them once; the platform handles packaging, dependencies, isolation, and execution.

Each section below goes deeper on what that means for see every run, every cost, every error.

View the full example in the docs →

[03 · Deep dive]

What makes it work.

📊

Cost per run

Computed from model cost config. Per-provider, per-token.

🎬

SSE streaming

Live logs as the agent runs. Backed by PostgreSQL LISTEN/NOTIFY.

🔍

Tool call trace

Every tool invocation with input, output, latency.

📤

Export on Enterprise

Ship logs to Datadog, Splunk, Grafana Loki.


Ship agents you can debug.

Full trace per run. Structured logs. No black boxes.