R
Get Routism

Self-hosted

Conduct many models
as one API

Self-hosted, OpenAI-compatible multi-model orchestration for coding agents.

Get RoutismSee install

./install.sh → routism · model routism-ultra

Story · 01

Agents speak one model. We conduct many.

Coding agents speak OpenAI Chat Completions. Routism is that endpoint — then a Conductor plans, fans work across your models, scores, and merges one answer.

  • 01Plan
  • 02Execute
  • 03Merge

conductor pipeline

  1. 01
    step
    Plan

    Conductor breaks the task

  2. 02
    step
    Execute

    Workers run in your pool

  3. 03
    step
    Merge

    One scored answer

Architecture · 02

Engine ≠ Workers

Docker runs Routism. Ollama runs the engine. You bring the workers.

Hidden engine models (planner / verifier) live on host Ollama. Your worker pool is BYOK — Ollama, MLX, Groq, or any OpenAI-compatible endpoint. The CLI wires OLLAMA_BASE_URL so Docker can reach host Ollama.

Enginehost · ollama
Conductor brains

eng-thinker · eng-verifier · eng-judge2 via Ollama — not in the Docker image, not selectable as workers.

eng-thinker
eng-verifier
eng-judge2
// not selectable as workers
Workersbyok · ≤5
Your pool (≤5)

Connect providers in the dashboard. Secrets stay on your machine.

  • ollama / mlx
  • groq / openai-compat
  • dashboard · secrets local

docker runs routism · ollama runs the engine · you bring the workers

ConductorPool

ENGINE

WORKERS

Install

One CLI. Full stack.

No accounts, OAuth, or billing — single installation, MIT licensed. Run ./install.sh then bare routism for interactive setup — Docker, Ollama, engine models, API + UI.

setup.sh
$ ./install.sh
$ routism
# Dashboard http://localhost:3000 · API http://localhost:8000/v1

# Agent env
export OPENAI_BASE_URL="http://localhost:8000/v1"
export OPENAI_API_KEY="rtm_…"
# model: routism-ultra
# Client timeout: 300–600 seconds
# ready — model routism-ultra

Dashboard http://localhost:3000 · API http://localhost:8000/v1

Why Routism

Built for agents, not demos

OpenAI-compatible

Point Hermes, Cursor, or any SDK at http://localhost:8000/v1. Model id: routism-ultra.

Self-hosted

Your machine, your Docker stack. No cloud lock-in for the control plane.

No SaaS accounts

No accounts, OAuth, or billing — single installation, MIT licensed.

Engine ≠ Workers

Docker runs Routism. Ollama runs the engine. You bring the workers.

Docs: OpenAI compat · API reference

Capabilities

  • OpenAI-compatible APIChat Completions endpoint
  • Self-hosted Docker stackControl plane on your machine
  • No SaaS accountsNo OAuth or billing required
  • Engine ≠ WorkersOllama engine · BYOK pool
  • Model idroutism-ultra

Works with your stack

Hermes
Cursor
OpenAI SDKs
Docker
Continue
Docker

compose stack

CLI

./install.sh

Local

your machine

Get Routism

Ready to host.
Ready to run.

Clone the public MIT repo. Install the CLI. Conduct your models.

Star & clone on GitHub
git clone https://github.com/Dreamstick9/Routism.git Routism
cd Routism
./install.sh
routism

Agent model id: routism-ultra · env: OPENAI_BASE_URL=http://localhost:8000/v1

export OPENAI_BASE_URL="http://localhost:8000/v1" export OPENAI_API_KEY="rtm_…" # model: routism-ultra # Client timeout: 300–600 seconds