Skip to content

The fastest, cheapest, most efficient open-source agent harness. Run more than 500 agents on a $10 VPS.

1.3 KB
Raw Blame

host_runtime

OpenHuman-owned bindings around tinyinference-local.

The reusable runtime implementation lives in the tinyinference-local crate: endpoint resolution and probing for Ollama / LM Studio / MLX / OMLX / OpenAI-compatible runtimes, status, prompts, vision, and embeddings. The user installs and runs that runtime and pulls its models; neither crate downloads, installs, or spawns anything. This directory contains only OpenHuman policy and integration seams:

  • core.rs: process-wide service ownership.
  • ops.rs and ops/: RPC operations, prompt access checks, agent dispatch, scheduler-gate acquisition, and temporary-file handling.
  • schemas.rs: OpenHuman controller registration and wire compatibility.
  • service/speech.rs: OpenHuman STT credential routing and Piper output.
  • service/transcription.rs: channel-facing transcription adapter shape.

Code here must call TinyInference directly. Do not copy local-runtime behavior back into this host module.

Further reading