slb350

Octoroute โ€” AI skill for Claude Code

AI community

Smart HTTP router for local LLMs (Ollama, LM Studio, llama.cpp).

How to install Octoroute

This entry records only its repository, not the path inside it, so there is no exact command to give. Open slb350/octoroute and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Octoroute does

Smart HTTP router for local LLMs (Ollama, LM Studio, llama.cpp). Rule-based + LLM-powered routing, health checks, load balancing, Prometheus metrics. Rust-native, zero-overhead.

Alternatives in AI

  • Repomix โ€” ๐Ÿ“ฆ Repomix is a powerful tool that packs your entire repository into a single, AI-friendly file 22.7k โ˜…
  • Ccstatusline โ€” by sirmalloc - A highly customizable status line formatter for Claude Code CLI that displays model info, git b 5.5k โ˜…
  • Inference Serving โ€” ai-research-skills vLLM, SGLang, TensorRT-LLM, llama.cpp 5.4k โ˜…

README

Octoroute

Octoroute v3 is an OpenAI-compatible inference fabric. Clients use stable virtual model names while Octoroute selects an eligible local llama.cpp pool or an ordered provider target.

OpenAI-compatible client
          |
          v
     Octoroute v3
       /      \
local pools   provider chain

The routing policy comes from configuration alone; Octoroute never classifies a prompt to choose a route. `X-Octoroute-Privacy: local-only` narrows a route before admission, so a local-only request cannot resolve provider credentials or disclose its prompt to a provider.

Current runtime

  • POST /v1/chat/completions with schema-preserving request forwarding.
  • GET /v1/models, liveness, readiness, and Prometheus endpoints.
  • Named virtual routes with ordered local-pool and provider steps.
  • Exact local context/capability checks and least-loaded member selection.
  • Per-member, per-provider, and inbound concurrency limits, plus an inbound per-minute request rate limit.
  • Lazy, isolated provider credentials from environment variables or bounded argv commands.
  • OpenAI-compatible HTTP dispatch for z.ai, OpenRouter, direct OpenAI, and similarly shaped endpoints.
  • Explicit Anthropic Messages translation for text, tools, reasoning, non-streaming responses, and incremental SSE.
  • Locked-down Codex CLI dispatch with ChatGPT-managed authentication, an allowlisted child environment, ephemeral read-only execution, and bounded structured output.
  • An explicit OpenRouter Auto profile owned by Octoroute.
  • Cached, bounded provider authentication/reachability probes and fixed-label Prometheus metrics for pool and provider admissions, responses, fallbacks, probes, and routing duration.
  • Closed fallback triggers and a held first byte, preventing target changes after response commitment; an optional first_byte_timeout_ms bounds how long a hung upstream holds its permits before the route falls forward.

Quick start

Requirements:

  • Rust 1.90 or ne