Tracing
- What happened?
- Which tools ran?
- Which model calls occurred?
- What was the timeline?
OPEN-SOURCE EXECUTION INTELLIGENCE
Tselora
Execution Intelligence for AI Agents
Tselora is an open-source execution intelligence layer that reconstructs AI-agent executions into a deterministic execution model—so teams can understand, reproduce, compare, and learn from agent behavior across runs.
Agent execution → Explorer → topology → Why? → Experience → Compare → Learn
From Agent Traces to Execution Intelligence
Traditional tracing gives visibility into individual executions. Tselora builds a deterministic execution model that makes execution history useful across runs. It does not replace OpenTelemetry, agent frameworks, or existing tracing systems. It is a layer that can work alongside them.
Tselora
- What was the execution structure?
- Which execution instances occurred?
- What structured decisions were emitted?
- What can be preserved as an Experience?
- How does one execution compare with another?
- Which historical executions share the same structure?
One Execution Model. Multiple Ways to Understand It.
Event history is authoritative. Tselora reconstructs execution state from those events. A projection engine provides the semantic execution model. Higher-level capabilities—Experience, reproduction lineage, comparison, and Learn—operate on that model.
Agent
Execution Events
Event Store
Projection Engine
Execution Model
Experience / Reproduction / Comparison / Learn
Capability journey
Observe → Explain → Control → Remember → Reproduce → Compare → Interpret → Learn
OBSERVE
Reconstruct agent execution topology and lifecycle from execution events.
EXPLAIN
Expose structured decision information emitted by the execution.
CONTROL
Provide cooperative runtime controls such as pause, resume, and stop.
REMEMBER
Create privacy-bounded Execution Experiences from terminal executions.
REPRODUCE
Record checkpoint and fork lineage while leaving application state restoration to the application.
COMPARE
Deterministically compare reconstructed executions and surface observable differences.
INTERPRET
Present execution differences and lineage through the Explorer.
LEARN
Retrieve historical executions sharing the exact same structure fingerprint.
Turn Executions into Experiences
An Execution Experience is a deterministic, privacy-bounded representation of a terminal Tselora execution. It is designed to preserve useful execution knowledge without becoming a second event log or a repository for prompts, completions, or chain-of-thought. Tselora does not store hidden model reasoning.
Execution Events
Projection
Terminal Execution
Execution Experience
Historical Retrieval
Preserve Execution Lineage
Tselora can record checkpoint and fork relationships between executions.
The application owns state restoration. Tselora owns the execution lineage.
Parent Run
Checkpoint
Fork
Child Run
Compare Executions, Not Just Traces
Tselora can reconstruct two executions and compare observable execution state. Comparison reports differences. It does not rank executions.
- Outcome
- Structured decisions
- Execution topology
- Timing/count information
- Controls
- Lineage
- Experience information when available
Find Executions with the Same Structure
Learn v0: a completed execution can be used as an anchor to retrieve historical Experiences with the exact same structure fingerprint. Results form a historical execution cohort. This is exact-match retrieval, not similarity ranking.
ANCHOR
run_123
SAME STRUCTURE
- run_098
- run_087
- run_054
- run_021
Architecture
Designed as a protocol-oriented execution intelligence layer that sits beside the agent you already run.
AI Agent
Execution Events
Event Store
Projection Engine
Execution Model
Experience
Compare
Reproduce
Learn
Supported agent stacks
Designed to sit alongside the agent frameworks you already use. Current adapters are first-slice integrations and are not zero-touch.
- OpenAI Agents SDK
- LangGraph
- CrewAI
- OpenTelemetry
Built in the Open
Tselora is open source and designed as a protocol-oriented execution intelligence layer. The Python package is published as tselora.
Built for Teams Building Agent Systems
- Agent developers
- AI platform teams
- Engineering teams operating multi-step agents
- Teams building multi-agent workflows
- Teams that need execution history beyond individual traces
Build agents you can understand.
Explore Tselora on GitHub and see how execution intelligence fits alongside your existing agent stack.
Explore Tselora on GitHubProblems this page addresses