Skip to content

diegosouzapw/OmniRoute: Claude Code Skills for an AI Gateway

By Flowy · Updated 2026-08-28

OmniRoute is an open source AI gateway that puts one endpoint in front of hundreds of model providers, with automatic fallback and prompt compression. This Claude Code plugin does not run the gateway itself. It packages 46 skills that configure, route, monitor, and back up a running OmniRoute install through its REST API or command line tool.

What is OmniRoute, and what does this plugin add to Claude Code

diegosouzapw/OmniRoute (57,292 GitHub stars) is an open source AI gateway: one endpoint in front of many model providers, so a coding tool sends one request and OmniRoute picks the provider and model that answer it. Its own project summary claims 350 providers, more than 1200 models, automatic fallback when a provider's quota runs out, and support for Claude Code, Codex, Cursor, OpenCode, Cline, and Copilot.

Flowy indexes a narrower slice: 46 skills, every one of them a skill, with no agents, commands, MCP servers, or hooks in this listing. These do not run the gateway, they administer one: an agent inside Claude Code uses them to configure and operate an OmniRoute install through its REST API or its command line tool. Most of the 46 come in matched pairs, one calling the API and one running from the terminal.

Which skills manage providers, models, keys, and settings

omni-providers and cli-providers list, add, test, and remove provider connections, OAuth flows, and API keys; the omni-providers description puts its own catalog at 327 providers, just under the project's 350. omni-models and cli-models browse the model catalog and resolve model name aliases. omni-proxies sets HTTP, HTTPS, or SOCKS proxies for outbound provider requests. omni-auth configures how OmniRoute authenticates incoming requests, and omni-settings holds the rest of its global configuration: system prompts, IP filters, and whether login is required. omni-api-keys and cli-keys manage OmniRoute's own API keys, the ones gating access to its endpoints, each with its own scopes and spending limits.

Which skills handle routing, budgets, and resilience

omni-combos-routing and cli-routing build routing combos out of 19 strategies, priority, weighted, and round robin among them, plus their fallback chains. omni-budget sets spending limits, token quotas, and rate limit policies per key or across an install. omni-resilience and cli-resilience watch circuit breaker states and latency, and can reset a provider stuck in a bad state. cli-health adds a live terminal dashboard over that same data.

Which skills compress prompts and manage the cache

omni-compression and cli-compression configure RTK and Caveman, the two compression modes behind the project's token savings, plus a stacked mode that combines them. omni-context-rtk and cli-contexts manage the filter sets and context relay settings those modes read. omni-cache is separate: it manages the LLM response cache itself, clearing entries, setting TTL policies, and tuning how similar two prompts must be to share a cached answer.

Which skills track usage, webhooks, backups, and tunnels

omni-usage-logs and cli-cost-usage filter and export call logs and cost breakdowns by provider, model, or date. cli-policy-audit covers compliance: audit logs, access policies, and request history. omni-webhooks registers endpoints for events like a completed request or an exceeded quota. omni-db-backups and cli-backup-sync back up and restore OmniRoute's SQLite database, while omni-sync-cloud syncs configuration and provider connections to cloud storage instead. omni-tunnels and cli-tunnel expose a local install to the internet over ngrok, Cloudflare, or a custom tunnel, for remote or CI access.

Which skills cover MCP, agent-to-agent requests, and the server itself

omni-inference is the integration surface an agent calls for an answer: OpenAI compatible chat, embeddings, images, audio, and the Responses API. cli-chat sends the same kind of request from a terminal, cli-batches runs batch inference jobs, and cli-eval runs evaluation suites and compares models. omni-mcp and cli-mcp cover OmniRoute's own MCP server and its permission scopes, and omni-agents-a2a and cli-a2a cover its agent-to-agent protocol, with built-in smart routing, quota management, provider discovery, cost analysis, and health reporting. Running the server itself is a smaller family: cli-serve starts and stops it, cli-setup handles first run configuration, and omni-version-manager installs and updates embedded background services such as 9Router and CLIProxyAPI.

Which skills manage OmniRoute's own skills, and the one that is not about OmniRoute at all

omni-cli-tools manages the CLI tool integrations exposed through OmniRoute's API. cli-plugins-skills covers the same ground from the terminal, plus two more concepts: installing and testing what its description calls Omni Skills, and managing OmniRoute's persistent memory. omni-github-skills and cli-skill-collector search GitHub for agent skill files such as SKILL.md or CLAUDE.md, score them for relevance, check for malware, and install matches into whichever coding tool they detect, Claude Code included. config-codex-cli does one specific version of that by hand, writing config.toml and seven profiles so Codex calls OmniRoute instead of OpenAI directly. Last, ponytail has nothing to do with OmniRoute: a separate, externally licensed skill pulled in from a different GitHub repository, pushing an agent toward the smallest solution that works.

How do I install it

Flowy indexes diegosouzapw/OmniRoute from its public GitHub repository. Install it from the listing page linked at the top of this guide, and all 46 skills above become available in that session.

Common questions

What is the diegosouzapw/OmniRoute Claude Code plugin?
It packages 46 skills for operating an already running OmniRoute install, the open source AI gateway from the same repository. Examples include `omni-providers` and `cli-providers` for managing model provider connections, `omni-combos-routing` for configuring routing and fallback strategies, and `omni-mcp` for OmniRoute's own MCP server. There are no agents, commands, MCP servers, or hooks in this listing, only skills.
Does this plugin include the OmniRoute gateway itself, or only the Claude Code skills?
Only the skills. Flowy indexes the 46 Claude Code skills that live in the `diegosouzapw/OmniRoute` repository for configuring and operating a gateway, not a copy of the gateway software. You still need an OmniRoute install running somewhere, local or remote, for skills like `omni-inference` or `cli-serve` to have anything to talk to.
How does OmniRoute find and install new agent skills automatically?
Two components handle that: `omni-github-skills` searches GitHub for repositories carrying `SKILL.md`, `CLAUDE.md`, or similar agent skill files, scores them for relevance, and checks for malware, while `cli-skill-collector` first detects which coding tools you have installed, `Claude Code` among them, then searches for matching skills and installs them through OmniRoute's own APIs. `cli-plugins-skills` is a third piece, managing what OmniRoute calls Omni Skills once they are already installed.
Can I use this plugin to point other coding tools, like Codex, at OmniRoute?
Partly. `config-codex-cli` is a step by step skill that configures OpenAI's Codex CLI specifically, detecting the operating system and shell and writing the config file and profiles Codex needs to call OmniRoute instead of OpenAI directly. The rest of the 46 skills are written for `Claude Code` sessions. The project's own summary says OmniRoute also works with Cursor, OpenCode, Cline, and Copilot, but this plugin does not ship a separate setup skill for each of those the way it does for Codex.
How many providers and models does OmniRoute actually support?
That depends on which number you read. The project's own summary claims 350 providers and more than 1200 models, while the `omni-providers` skill's own description puts the provider catalog at 327. Both figures come from the same project, so treat either as approximate, and use the `omni-providers` or `cli-providers` skill to see what is actually configured on your own install.