capsule-trace 0.1.1

Deterministic replay & time-travel debugger for AI agents

capsule-trace

Deterministic replay & time-travel debugger for AI agents.

pip install capsule-trace

---

Quick Start

import capsule_trace

@capsule.trace(agent_name="my-agent")
def run_agent(query: str) -> str:
    from openai import OpenAI
    client = OpenAI()
    response = client.chat.completions.create(
        model="gpt-4o",
        messages=[{"role": "user", "content": query}],
    )
    return response.choices[0].message.content

That's it. Every LLM call is now captured automatically.

CLI

# List captured sessions
capsule-trace list

# Inspect a session step by step
capsule-trace show ses_01HXYZ123

# Export to shareable file
capsule-trace export ses_01HXYZ123 --output bug.capsule

# Replay deterministically
capsule-trace replay ses_01HXYZ123

# Branch from step 7 with different temperature
capsule-trace branch ses_01HXYZ123 --from-step 7 --modify temperature=0.0

# Diff two sessions
capsule-trace diff ses_01HXYZ ses_01HABC

Supported Providers

Provider	Status
OpenAI (sync + async)	✅ Sprint 2
Anthropic (sync + async)	✅ Sprint 2
Google Generative AI	✅ Sprint 2
LangChain	Sprint 4
LangGraph	Sprint 4

Disable at Runtime

CAPSULE_DISABLE=1 python my_agent.py

Zero overhead when disabled — checked once at function entry.

License

Apache 2.0 — see LICENSE.