Expand description
§TokenFuse
Runtime cost control & security for AI agents — a drop-in proxy between your agent and its LLM/tool providers that enforces per-run budgets, detects runaway loops, offers a kill-switch, and keeps secrets out of the model’s context.
This is the project’s umbrella crate. TokenFuse runs as a service, not a
library dependency: the gateway ships as the tokenfuse binary and as Docker
images on GHCR.
docker run -p 4100:4100 -e TOKENFUSE_MODE=enforce ghcr.io/taipanbox/tokenfusePoint your provider client at http://127.0.0.1:4100 and attach a few
X-Fuse-* headers. See the repository for the full documentation, the HA
cluster, the hosted Cloud, and the MCP credential-broker.
Constants§
- DEFAULT_
GATEWAY - Default gateway address the proxy listens on.
- VERSION
- The released TokenFuse version this crate documents.