Integrations

omp (oh-my-pi)

Run open-weight, GPT and Claude models in omp through Roteo, all on one key.

omp (oh-my-pi) is a terminal coding agent. It reaches Roteo three ways, all with the same key:

You wantUseModels
Open-weighta custom provider in omp's models.ymldeepseek-v4.1-flash-1017, glm-5.3 and more
GPTa second custom provider, on the Responses wiregpt-6.1-sol, gpt-6.1-terra and more
ClaudeRoteo's omp plugin, which runs Claude Code underneathclaude-opus-5-5, claude-sonnet-5-5, claude-haiku-5-5, claude-fable-5-2

Install omp

curl -fsSL https://omp.sh/install | sh
bash · 1 lines

Open a new terminal afterwards so omp is on your PATH.

Set your key

All three read your Roteo key from ROTEO_API_KEY, in the environment omp starts from:

export ROTEO_API_KEY="sk_your_key"
bash · 1 lines

On macOS and Linux, add the line to your shell profile (~/.zshrc, ~/.bashrc) to keep it. On Windows, setx keeps it for good: open a new terminal before starting omp.

Open-weight and GPT models

Add the providers

Create ~/.omp/agent/models.yml (or add to it) with both providers:

providers:
  roteo:
    baseUrl: https://api.roteo.ai/v1
    api: openai-completions
    apiKey: ROTEO_API_KEY
    models:
      - id: deepseek-v4.1-flash-1017
        name: DeepSeek V4.1 Flash
        contextWindow: 1048576
        maxTokens: 32000
      - id: glm-5.3-flash
        name: GLM 5.3 Flash
        contextWindow: 1048576
        maxTokens: 131072
      - id: glm-5.3
        name: GLM 5.3
        contextWindow: 262144
        maxTokens: 131072
      - id: kimi-k3
        name: Kimi K3
        contextWindow: 1048576
        maxTokens: 32000
      - id: qwen3.8-max
        name: Qwen 3.8 Max
        contextWindow: 1000000
        maxTokens: 32000
  roteo-gpt:
    baseUrl: https://api.roteo.ai/v1
    api: openai-responses
    apiKey: ROTEO_API_KEY
    models:
      - id: gpt-6.1-sol
        name: GPT-6.1 Sol
        reasoning: true
        contextWindow: 1000000
        maxTokens: 128000
      - id: gpt-6.1-terra
        name: GPT-6.1 Terra
        reasoning: true
        contextWindow: 1000000
        maxTokens: 128000
      - id: gpt-6-astra
        name: GPT-6 Astra
        reasoning: true
        contextWindow: 1000000
        maxTokens: 128000
      - id: gpt-6-luna
        name: GPT-6 Luna
        reasoning: true
        contextWindow: 1000000
        maxTokens: 128000

apiKey: ROTEO_API_KEY is the name of the environment variable, so the key itself never sits in the file. Spell it exactly: omp sends a name it can't find as a literal key, and the request fails with a 401.

GPT runs on its own provider because it uses the Responses wire (openai-responses), the same one Codex uses. Both providers share the base URL and the key.

Check that omp sees them:

omp models roteo
omp models roteo-gpt
bash · 2 lines

Pick a model

On its first start, omp opens a setup wizard that asks you to sign in to a provider. Press Esc to skip that step: Roteo comes from models.yml, not from the sign-in list. Then choose a roteo/... or roteo-gpt/... model as your default. Later, /model switches models in a session.

You can also set the defaults in ~/.omp/agent/config.yml:

modelRoles:
  default: roteo-gpt/gpt-6.1-sol
  smol: roteo/deepseek-v4.1-flash-1017
yaml · 3 lines

omp's smol role handles quick side jobs, such as naming the session. Point it at a fast, cheap model such as deepseek-v4.1-flash-1017 so they don't run on a pricier main model. When your default is already DeepSeek, leave smol out: omp then uses the default for them too.

Models

Use any open-weight id from Models under roteo, and any gpt-* id under roteo-gpt. To add one later, put it in the provider's models list and restart omp. Claude ids are refused on these two providers; for Claude, use the plugin below.

If it fails

You seeDo this
"401 Invalid API key"Check the key, and that apiKey names ROTEO_API_KEY exactly
"402 Insufficient balance..."Top up in your dashboard. The message says how much is needed. Promo credit covers Claude and GPT only, so open-weight models need wallet balance
A model is missing from /modelRestart omp after editing models.yml
"Third-party agent frameworks aren't supported on Claude models"A Claude id is under roteo. Use a roteo-claude/... model from the plugin
"Only GPT models are supported on /v1/responses..."A model that isn't GPT is under roteo-gpt. Put open-weight ids under roteo, and use the plugin for Claude

Claude models

The roteo-claude plugin runs the official Claude Code as omp's model client. omp keeps its own interface, tools and approvals; Claude Code sends each request to Roteo with your key. No Claude login is needed, and Claude Code comes with the plugin.

1. Requirements

  • omp 18.20 or newer. Check with omp --version.
  • git and Node.js with npm, to download the plugin and install its packages.
  • ROTEO_API_KEY set as above.

2. Install the plugin

omp loads every plugin in ~/.omp/agent/extensions/ when it starts:

git clone https://github.com/roteo-ai/omp-plugin-roteo ~/.omp/agent/extensions/omp-plugin-roteo
cd ~/.omp/agent/extensions/omp-plugin-roteo
npm ci --omit=dev --ignore-scripts
bash · 3 lines

Check that omp loads it with omp models roteo-claude.

3. Pick a model

In omp, run /model and choose one:

ModelRuns
roteo-claude/claude-opus-5-5Claude Opus 5.5
roteo-claude/claude-sonnet-5-5Claude Sonnet 5.5
roteo-claude/claude-haiku-5-5Claude Haiku 5.5
roteo-claude/claude-fable-5-2Claude Fable 5.2
roteo-claude/claude-fable-5-1Claude Fable 5.1

All of them run with a 1M-token context window. To make one the default, set default: roteo-claude/claude-sonnet-5-5 under modelRoles in ~/.omp/agent/config.yml.

Every request is billed to your Roteo balance. omp's own cost estimate shows $0 for these models, so check each request's charge in your dashboard. As everywhere else, Fable's own safeguards turn down cybersecurity-related requests, so use Opus or Sonnet for those.

The plugin gives Claude Code its own folder (~/.omp/agent/roteo-claude/claude-config) and loads none of your Claude Code settings, hooks or login, so your own Claude Code stays as it is. Claude Code's telemetry and error reports stay off.

If it fails

You seeDo this
"No Roteo API key. Set ROTEO_API_KEY..."Set the key in the environment omp starts from, then restart omp
"Failed to authenticate. API Error: 401 Invalid API key"The key is wrong or revoked. Check it in your dashboard
"API Error: 402 Insufficient balance..."Top up in your dashboard. The message says how much is needed
No roteo-claude models in /modelThe plugin must sit in ~/.omp/agent/extensions/omp-plugin-roteo, with npm ci run inside it. Then restart omp
"No API key for provider: roteo-claude" after a plan-mode stepRestart omp
"Fable 5.2's safeguards flagged this message"Fable turns down cybersecurity-related requests, in every tool. Use Opus or Sonnet for that task

To update the plugin, run git pull and npm ci --omit=dev --ignore-scripts in its folder, then restart omp. Its settings and source are in the README.

omp (oh-my-pi) | Roteo