More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Free LLM Chain: AI features at $0 that keep answering

    3

    Chain free-tier LLMs behind one call: fallback rules, first-text deadline, daily caps, lite routing, BYOK. Stdlib client + offline tests.

    $9

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Free LLM Chain: AI features at $0 that keep answering

    Free LLM Chain: AI features at $0 that keep answering

    Example session with this skill installed

    My app's free AI assistant sometimes hangs for minutes, and one user burned the whole quota. Make it robust and still free.

    • Read your context and instructions
    • Compiled the free llm chain

    Run 2026-11-16 (offline, mock server):

    $ python3 -I -m unittest discover -s tests -v
    ... 31 tests, Ran 31 tests in 6.0s, OK

    $ env -i python3 -I scripts/chain_client.py hi
    error: no provider key set. Export at least one of GROQ_API_KEY, OPENROUTER_API_KEY, GEMINI_API_KEY (Gemini also needs CHAIN_ORDER to name it). Refusing to call anything. (exit 2)

    $ chain_client.py "Capital of France?"
    {"attempt": 0, "backend": "groq", "error": "no text after 1 s", "event": "backend_failed", "kind": "slow"}
    {"attempt": 1, "backend": "openrouter", "event": "answered", "first_ms": 1}
    Paris.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Free tiers rate-limit, retire models without notice, go paid-only and stall: the model bill is zero but the feature stops answering on a Thursday. This skill is the design that keeps an AI feature answering at $0, plus the guards that stop one user draining the shared quota.

    What it does

    Your agent builds a chain of free providers (Groq, OpenRouter free models, optionally Gemini's free tier) behind one OpenAI-compatible call. It hops on 429, 5xx, timeouts and empty 200s, never splices two models' text, cuts a reasoning model that streams frames but no words (a first-text deadline), replaces retired model slugs from the provider's own /models list, caps each account per day with a refund when nothing answered, routes easy turns to a lite model, and sets up prefix caching correctly. It also covers keyless default versus bring-your-own-key, key expiry and the prefs write race that loses keys.

    What you get

    PathHolds
    SKILL.md8 rules, error-class table, deadlines, caps, routing, method
    references/chain-design.mdhop rules, 200-with-no-text, TPM 413, tool loops, runaway answers
    references/budgets-and-keys.mdaccount, trial and global walls, spend/refund, BYOK
    references/routing-and-caching.mdlite routing, thinking level, implicit cache, answer cache
    references/operations.mddead-chain signatures, latency, what to log, privacy, deploy checklist
    scripts/chain_client.pystdlib chain client: CLI and importable, JSON log of who answered
    tests/test_chain_client.py31 offline tests on a local mock server
    templates/chain-config.mdenv block and settings table

    Who it's for

    Developers shipping an assistant to users who will not paste an API key, on a budget of zero, and anyone whose free-tier integration hangs for minutes before the first word.

    Scope

    Covered: transport and fallback semantics, deadlines, caps and refunds, BYOK, lite routing, prefix caching, logging and alerting, deploy checklist. Figures from a real deployment are dated and labelled observed.

    Not covered: paid-tier tuning, self-hosted models, embeddings, vision. Free-tier limits and model names change often, so the package never states them as fact: it says check the provider page. Tested offline against a mock server; no real provider was called, so run --probe with your own keys first.

    Not legal advice on provider terms.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Workflow Automation.

    Tested offline against a mock server only; no real provider was called. Default model names and the Gemini base URL are starting points to confirm with --models. Free-tier limits and slugs change often; the package states none as fact. The script's daily cap is per account in a local file; a multi-process server needs its own shared counter.

    How to use

    1. 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. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before buying a skill.

    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 1 day ago

    Accounts
    GroqGoogle AIOpenrouter
    Software
    Python 3.9+A free-tier LLM provider key

    Standard library only. A Groq, OpenRouter or Gemini key is needed to run for real; the offline tests need none.

    Permissions

    Allowed Hosts

    api.groq.com
    openrouter.ai
    generativelanguage.googleapis.com

    The script calls only the providers you set keys for; keys are read from environment variables and never stored or logged.

    Listed1 day ago

    What's inside

    Frequently Asked Questions

    No. They change often, so the skill says check the provider page and keeps numbers in config. Figures from a real deployment are dated and labelled observed.

    No. The 31 tests run offline against a local mock server. Run chain_client.py --probe with your own keys, from the machine that will serve users, before relying on it.

    Opt-in only: name it in CHAIN_ORDER. Read its free-tier data terms first, since free traffic may be used for training.

    No. The client is Python 3.9+ standard library only and speaks the OpenAI chat-completions wire format; the design applies to any language.

    Per account via a JSON state file with --account, --state and --daily-cap; for a server you persist the counter in your own store using the same spend-before, refund-on-failure rule.

    New in Workflow Automation