Capsule
Capsule provides deterministic replay and time-travel debugging for AI agents.
FreeCoding & Dev
Visit website Free to browse · updated 2026-06-23
What is Capsule?
The tool wraps agent code with a simple decorator to automatically capture all relevant events during execution. Sessions can then be replayed deterministically from stored data, allowing inspection of the exact world state at any step with millisecond precision. Additional capabilities include forking sessions at chosen points to experiment with different prompts or models, along with automatic tagging for common failure modes. Cost tracking per session and support for multiple frameworks ensure broad applicability in development workflows. Built-in compliance features support audit requirements through signed records and options for data redaction or self-hosted deployment, making it suitable for regulated environments where reproducibility is essential.
Key features
Deterministic capture of every LLM call, tool use, and state change
Time-travel replay from stored cassettes with bit-exact output
Branch and fork sessions at any step to swap prompts or models
Export portable .capsule files with SHA-256 verification
Per-session token cost analytics and latency tracking
Framework-agnostic support for LangChain, LlamaIndex, AutoGen and raw LLM APIs
Compliance logging for EU AI Act with optional GDPR PII redaction
What you can use Capsule for
Debugging AI Agent Failures
Capture every LLM call, tool use, and state change during agent execution, then rewind sessions with deterministic replay to inspect exact states at the point of failure without live API calls.
Sharing Reproducible Bug Reports
Export portable .capsule files containing all events, cassettes, and snapshots that can be attached to bug reports for bit-exact reproduction on any machine.
Experimenting with Agent Variations
Fork sessions at any step to swap prompts, models, or tool responses, then run branches live and compare outputs side-by-side for iterative improvements.
How to use Capsule
- 1Install via pip install capsule-trace
- 2Wrap your agent with the @capsule.trace decorator
- 3Execute the agent to automatically capture sessions
- 4Use the CLI to replay a session with capsule-trace replay
- 5Branch at a chosen step or export the .capsule file
Capsule pricing
Pricing model: Free. Plan details are indicative — check the site for current prices.
Free
Free
- Capture & replay
- Deterministic replay
- Self-host available
- SOC 2 in progress
Editor's verdict
Pros
- +Single decorator for zero-code-change capture
- +Open, self-describing .capsule format with cryptographic integrity
- +Full CLI for replay, branch, diff and export operations
Cons
- –SOC 2 certification still in progress
- –Python-only SDK at present
- –Self-hosting required for air-gapped or VPC deployments
Our take: Capsule is a solid coding & dev choice. It's valued for single decorator for zero-code-change capture and open, self-describing .capsule format with cryptographic integrity. The main trade-off is soc 2 certification still in progress. A good pick if you want capable AI without a high upfront cost.
Frequently asked questions
A single compressed, self-describing archive that holds every event, cassette, and snapshot from an agent session, with SHA-256 hashes for verifiable integrity and offline replay.
Summary
Capsule is a solid coding & dev choice. It's valued for single decorator for zero-code-change capture and open, self-describing .capsule format with cryptographic integrity. The main trade-off is soc 2 certification still in progress. A good pick if you want capable AI without a high upfront cost.
Did you find this helpful?
User reviews
Verified reviews from the community shape this tool's rating.
Loading reviews…
Capsule alternatives
Similar coding & dev tools worth comparing.
