More screenshots
Works with the AI Agents you already use
Route Each Task to the Right Model
Your agent stops using one model for everything — easy work goes to a free local model, hard work escalates only when a check fails.
$12
Route Each Task to the Right Model
Example session with this skill installed
Explain how to set up a Cascade tier that uses a local model for drafts and escalates to a hosted model if the response is too short or lacks a code block.
- Read your context and instructions
- Compiled the route each task
To implement a Cascade tier, you must define the models in your modelrouter.yaml and set the escalation triggers. The router evaluates the first draft against your configured quality gates before deciding whether to call the stronger model.
Cascade Configuration Example
models:
- id: llama3-local
endpoint: http://localhost:11434/v1
price_input: 0.0
price_output: 0.0
- id: gpt-4o
endpoint: https://api.openai.com/v1
key_env: OPENAI_API_KEY
price_input: 5.0
price_output: 15.0
tiers:
cascade:
draft: llama3-local
strong: gpt-4o
gates:
min_length: 50
require_code: true
refusal_patterns:
- "as an ai"
- "i cannot"
policies:
budget: 10.0
The router will now process calls to model: "cascade". If llama3-local returns 30 characters or no triple-backtick code block, it automatically triggers a request to gpt-4o.
Next steps
- Run
python -m modelrouter serveto activate the endpoint. - Monitor
http://127.0.0.1:4000/to see real-time escalation reasons. - Test your gates using
python -m modelrouter route --explain "Write a python script" --tier cascade.
Connects securely to your tools. The creator never sees your data.
About this skill
Point an agent at one strong model and you pay strong-model prices to rename variables. Point it at one cheap model and hard problems get weak answers. This skill sets up modelrouter-ai, a free proxy I built, between your agent and your models. The agent sends one request; the router decides which model answers — a free local model, a cheap hosted one, or a strong one.
Inside the package: the skill file covering install and configuration, a filled example config, and a routing-patterns reference that maps task types to the model each deserves — boilerplate and renames stay local, everyday feature work escalates on failure, subtle debugging and security review go to the strong tier. The proxy itself is free and MIT-licensed on GitHub; this package is the setup, the routing judgment, and the cost controls.
A run goes like this: the router offers three patterns. Single — one model answers, no hidden fallback. Cascade — a cheap or local model drafts, simple quality checks judge the draft (empty answer, a refusal, code missing when code was asked for), and only a failed check escalates to the strong model, with the reason logged. Critique — a second model scores the draft and sends it back for revision below a score you set. Every call is priced from a table you control, and a daily budget cap warns at 80% and stops at 100%, so an agent loop can't burn a card overnight.
Streaming is non-incremental until the full response is gated. Binds to 127.0.0.1; requires manual auth for remote access. Quality gates may cause double billing on high-effort escalations.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 6 days ago
Requires