English documentation · runtime 7.24.4 · SDK 2.6.7. Content is maintained with runtime development; see each guide's scope and review date.
Free-Tier API Quickstart
Status: Active Scope: current-state Last reviewed: 2026-08-28 Owner: ax-code runtime
AX Code does not provide model quota. A “free” run uses quota or credits offered by an external model provider, whose availability, limits, data terms, and model catalog can change independently of AX Code.
Shortest supported path: Groq
Groq is a default AX Code provider, its current AX Code model catalog contains tool-capable models, and Groq publishes Free Plan limits for those model IDs.
- Create a Groq account and API key in the Groq Console.
- Store the key in AX Code:
ax-code providers login groq
- Confirm the models available in your installed AX Code release:
ax-code models groq
- Open
ax-code, run/connect, and choose Groq and one of the listed models. For a headless smoke test, the current bundled catalog supports:
ax-code run --model groq/openai/gpt-oss-120b "Inspect this repository without editing it and summarize its purpose."
If that exact model is not printed by ax-code models groq, use a model ID that is. The installed catalog is
authoritative for the installed release.
Built-in free-tier and free-credit options
These providers are in AX Code’s default setup picker. “Built in” means AX Code supplies the protocol adapter, model metadata, credential flow, and provider endpoint; it does not mean every model or request is free.
| AX Code provider | No-cost path | Important constraint |
|---|---|---|
groq |
Groq Free Plan | Rate limits are per model and account; check Groq’s live limits. |
google |
Free-tier Gemini API models | Only models marked free in Gemini API pricing qualify. Google states that free-tier content may be used to improve its products. |
unorouter |
Model IDs ending in :free |
UnoRouter is a third-party gateway. Its free-model limits are intentionally low and availability can fluctuate. |
huggingface |
Recurring Inference Providers credit | Free accounts currently receive a small monthly credit, subject to change; see Hugging Face pricing. A full agent task can exhaust it. |
zai |
Current GLM-5.3 SKUs on the PAYG API | AX Code’s picker starts at GLM-5.3. Connect the zai provider, not zai-coding-plan, and check Z.AI pricing for the current no-cost SKU. |
openrouter |
openrouter/free or a model-specific :free variant |
The provider preset is built in, but the AX Code curated snapshot does not currently list the free router. Add it explicitly as shown below. |
For UnoRouter, connect with ax-code providers login unorouter, run ax-code models unorouter, and choose a model
whose ID ends in :free. Do not assume the paid twin without that suffix is free. UnoRouter’s own catalog notes that
not every free model supports tool calling; AX Code only surfaces the tool-capable entries in its bundled catalog.
For Google or Hugging Face, connect with the matching provider ID and then use ax-code models google or
ax-code models huggingface. Check the provider’s live pricing page before choosing a model.
Add OpenRouter’s free-model router
OpenRouter documents openrouter/free as a zero-cost router that selects an available free model compatible with the
request. Because that router is not in AX Code’s current curated snapshot, add a conservative local model entry to
the ax-code.json at your project root (or merge it into your global AX Code config):
{
"$schema": "https://ax-code.app/docs-assets/schema/config.schema.json",
"provider": {
"openrouter": {
"models": {
"openrouter/free": {
"name": "OpenRouter Free Models Router",
"tool_call": true,
"limit": {
"context": 32768,
"output": 4096,
},
},
},
},
},
}
The limits above are conservative AX Code budgeting values, not an entitlement from OpenRouter. Then connect and verify the merged model catalog:
ax-code providers login openrouter
ax-code models openrouter
ax-code run --model openrouter/openrouter/free "Inspect this repository without editing it and identify its test command."
The repeated openrouter is intentional: the first segment is the AX Code provider ID and openrouter/free is the
model ID expected by OpenRouter. Free-router availability and selection vary by request; see
OpenRouter’s free-router documentation.
How to use the external free-API catalog
awesome-free-llm-apis is useful for discovering current offers, but it is not an AX Code compatibility or endorsement list:
- A provider in Supported Providers and Models has a default AX Code setup preset.
- A provider merely present in
models-snapshot.jsonis registry-backed, not necessarily curated for the default setup flow. - Another service can work through Custom and Gateway Providers when it exposes a compatible OpenAI or Anthropic endpoint.
- “OpenAI compatible” is necessary but not sufficient for a coding agent. The selected model also needs streaming text responses and reliable tool/function calling.
- Provider names, model IDs, free quotas, regions, and account requirements in third-party lists can become stale. Verify them against the provider’s official pricing, model, and rate-limit pages before relying on them.
At this page’s review date, the services named by that catalog map to AX Code as follows:
| AX Code relationship | Catalog services | Meaning |
|---|---|---|
| Default setup preset | Google Gemini (google), Groq (groq), OpenRouter (openrouter), Hugging Face (huggingface), UnoRouter (unorouter) |
Visible in the normal provider setup flow and covered above. |
| Registry-backed, not a default preset | Mistral (mistral), Cohere (cohere), Zhipu AI (zhipuai), AnyAPI (anyapi), Cerebras (cerebras), NVIDIA NIM (nvidia), Cloudflare Workers AI (cloudflare-workers-ai) |
Metadata and an adapter exist in the bundled registry, but the provider is not part of the curated default setup flow. Explicit configuration does not guarantee that a free model supports agent tools. |
| Custom protocol only | Api.Airforce, Kluster AI, LLM7.io, Pollinations AI | No AX Code preset. Configure the exact endpoint and model only if its current API is compatible with the custom provider contract. |
In particular, the github-copilot preset is not the github-models API listed by some catalogs. Do not substitute
one provider ID or credential flow for the other. The zai and zhipuai general APIs start at GLM-5.3; older GLM 3,
GLM 4, GLM-5, and GLM-5.2 SKUs stay filtered out.
Safe evaluation practices
- Test with a public or disposable repository first. A hosted provider or gateway receives the prompts and code context sent to its API.
- Enable
workspace-writeisolation, or useread-onlyfor a review-only smoke test. - Start with one short task and one agent. Council, arena, long sessions, and large repositories can consume many requests and quickly hit free-tier limits.
- Store credentials with
ax-code providers loginor an environment variable; never commit API keys toax-code.json. - Treat
429and temporary unavailability as normal free-tier constraints. Wait for the provider’s retry window or choose another currently listed free model.