ALL PROJECTS
Totem Runtime
AGENT EXECUTION
Totem Runtime Studio
An execution layer for AI agents. Permission before action. Evidence after.
THE IDEA
A model can propose an action. A runtime decides whether it should happen. Totem puts scheduling, capability checks and an audit trail between an agent and its tools.
HOW IT WORKS
- 01Plan an execution graph
- 02Evaluate capabilities
- 03Schedule independent branches
- 04Invoke the tool
- 05Record the audit chain
INSIDE THE BUILD
DAG scheduling uses Tokio to run independent work concurrently.
CAPSEM applies default-deny capability policies before tool invocation.
MCP connects the runtime to a discoverable tool registry.
BLAKE3-linked audit events make changes to execution history detectable.
Security belongs in the execution path, not only in the prompt.
THE TRADE-OFFS
Transport support and OS-level isolation have different maturity levels. A capability check is not a substitute for a complete operating-system sandbox.
BUILT WITH
- Rust
- Tokio
- Axum
- MCP
- BLAKE3