How the parts connect
Continue uses roles. Each model in config.yaml has one or more roles. Chat and agent mode use the chat role. Ghost-text completion uses the autocomplete role. If you do not set roles, the default is chat, edit, apply and summarize.
| Continue feature | FreeLLMAPI endpoint | Model needs |
|---|---|---|
| Chat, edit, apply | /v1/chat/completions | Any chat model |
| Agent mode | /v1/chat/completions with tools | Tool calling (tool_use capability) |
| Tab autocomplete | /v1/completions (legacy prompt and suffix) | A fast model |
FreeLLMAPI serves /v1/completions for editor autocomplete clients that send legacy OpenAI prompt and suffix requests. Continue sends these requests when you set useLegacyCompletionsEndpoint: true.
Step 1: Install FreeLLMAPI and run the setup command
Install on macOS or Linux
curl -fsSL https://freellmapi.co/install.sh | bashOpen http://localhost:3001. Add at least two free provider keys on the Keys page. Copy your unified key. Then run the setup command:
One command
npx freellmapi setup-continue --url http://localhost:3001 --api-key <unified-key> --dry-run
npx freellmapi setup-continue --url http://localhost:3001 --api-key <unified-key>The command makes a backup and merges a chat model into ~/.continue/config.yaml. It puts the key in ~/.continue/.env. The YAML refers to the key as a secret, so the raw key is not in the config file.
.env: the Continue docs say that the IDE extensions cannot read your shell environment variables. Continue looks for secrets in <workspace>/.env, <workspace>/.continue/.env, ~/.continue/.env and then the process environment.To test the connection without the IDE, use the Continue CLI. The command is cn, from the npm package @continuedev/cli. The Continue docs mark the CLI as alpha.
One-shot check
npm install -g @continuedev/cli
cn -p "Say hello"The VS Code and JetBrains extensions read the same ~/.continue/config.yaml. You configure the gateway one time for both IDEs. Each request goes to your computer, so localhost works and no tunnel is necessary.
Step 2: Add the chat and autocomplete models
Open config.yaml from the configs menu in Continue. Use the cog next to Local Config. The file is at ~/.continue/config.yaml on macOS and Linux, and at %USERPROFILE%\.continue\config.yaml on Windows. A complete file for chat, agent mode and autocomplete looks like this:
~/.continue/config.yaml
name: FreeLLMAPI
version: 1.0.0
schema: v1
models:
- name: FreeLLMAPI
provider: openai
model: auto
apiBase: http://localhost:3001/v1
apiKey: ${{ secrets.FREELLMAPI_API_KEY }}
capabilities:
- tool_use
defaultCompletionOptions:
contextLength: 128000
- name: FreeLLMAPI Autocomplete
provider: openai
model: auto
apiBase: http://localhost:3001/v1
apiKey: ${{ secrets.FREELLMAPI_API_KEY }}
useLegacyCompletionsEndpoint: true
roles:
- autocomplete~/.continue/.env
FREELLMAPI_API_KEY=<unified-key>capabilities: tool_useis required for agent mode. Continue usually detects it, but the gateway modelautois not a known model name, so set it.contextLengthtells Continue how much context to send. Set it to the context of your model.- Replace
autoin the autocomplete entry with a pinned model id if you want the same model for each keystroke.
Step 3: Choose models
For chat and agent mode, use a model with tool calling and a large context window. The table shows free models that meet both rules today.
| Model | Free on | Context | Free limits (best offer) | Tools | Vision |
|---|---|---|---|---|---|
| GLM 5.3 | UnoRouter | 1M | 1 req/min 1 req/min per model | Yes | No |
| GLM 5.3 Flash | NVIDIA NIM, UnoRouter | 1M | 40 req/min 40 RPM | Yes | Yes |
| Gemini 3.8 Flash | Google AI Studio | 1M | Recurring provider quota | Yes | Yes |
| Qwen3.8 Flash Next | AMD Radeon Cloud | 262K | Recurring daily account allowance | Yes | No |
| Qwen3.8 Flash | ElectronHub | 1M | $0.25/week shared recurring credit | Yes | Yes |
| Gemini 3.7 Flash | Google AI Studio | 1M | 10 req/min 20 req/day 250,000 tokens/min ~3M | Yes | Yes |
| GPT-5.6 Luna | ElectronHub | 1M | $0.25/week shared recurring credit | Yes | Yes |
| Dots3 Note Preview | Kilo Gateway, OpenRouter | 512K | 20 req/min 50 req/day 50 req/day account-wide (1,000 after a $10 purchase) | Yes | Yes |
Showing 8 of 109 matching free models. See all models.
Live data from the FreeLLMAPI catalog (version 2026.11.19, checked 19 November 2026). Free limits change often. The catalog is checked every day.
For autocomplete, the Continue docs recommend models that are trained for completion, for example Codestral and Qwen Coder. They say that large chat models and thinking models do not give useful completions. The table shows the free Codestral and Qwen Coder offers today. Copy the exact id from GET /v1/models.
| Model | Free on | Context | Free limits (best offer) | Tools | Vision |
|---|---|---|---|---|---|
| Codestral | DreamPrompting, LLM7, Mistral +1 more | 256K | 100 req/min 5,000 req/day 500,000 tokens/day 500K tokens/day · 5,000 req/day | Yes | No |
| Codestral 2508 | ACLIDE, Api.Airforce, ElectronHub +1 more | 256K | 1 req/min 1,000 req/day 1,000 req/day · 1 req/min | Yes | No |
| Qwen2.5 Coder 32B | Cloudflare Workers AI | 33K | ~10-20M | Yes | No |
| Qwen3 Coder 30B A3B Instruct | ElectronHub, ModelScope | 262K | 200 req/day Daily login: 200 credits + 50 for linked Alibaba Cloud; 24h expiry; shared; noncommercial experience | Yes | No |
| codestral-latest | Api.Airforce | 32K | 1 req/min 1,000 req/day 1,000 req/day · 1 req/min | No | No |
| Qwen3 Coder Plus | ACLIDE | - | €20/month shared internal usage credits · no card | Yes | No |
| Qwen3 Coder 480B A35B Instruct (turbo) | Inferbase | 8K | 5,000 requests/month; 100 requests and $0.50 usage/day shared | No | No |
| Qwen3 Coder | Septor Labs | 128K | Shared daily allowance | Yes | No |
Showing 8 of 9 matching free models. See all models.
Live data from the FreeLLMAPI catalog (version 2026.11.19, checked 19 November 2026). Free limits change often. The catalog is checked every day.
You can also use a routing model id in place of auto. For example, auto:smart prefers the models with the highest intelligence, and auto:fast prefers measured speed. Make a separate model entry for each one, then select it in the Continue model menu.
Autocomplete settings and limits
Autocomplete sends a request after almost every pause in typing. On free tiers, this is the fastest way to use your quota. Use the autocompleteOptions block to send fewer, smaller requests.
Add under the autocomplete model entry
autocompleteOptions:
debounceDelay: 500
maxPromptTokens: 1024
onlyMyCode: truedebounceDelayis the wait in milliseconds after you stop typing. A higher value sends fewer requests. The Continue docs example uses 250.maxPromptTokenslimits the size of each completion prompt.- Put autocomplete on a model with high per-minute limits. Do not put it on a model with 20 requests per day.
- Keep chat and autocomplete on different models. Then autocomplete does not use the quota that chat needs.
When a provider sends a 429 error, FreeLLMAPI tries the next model in the chain. The X-Routed-Via response header shows the provider that answered.
Troubleshooting
| Problem | Cause | Fix |
|---|---|---|
401 "Invalid API key" | The key is wrong, or Continue did not load the secret | Put FREELLMAPI_API_KEY in ~/.continue/.env, then reload the IDE |
| Agent mode is not available for the model | The model has no tool_use capability in the config | Add capabilities: [tool_use] to the chat model |
| No ghost text | No model has the autocomplete role | Add the autocomplete entry with roles: [autocomplete] |
| Completions are slow | A large model or a slow free tier | Pin a fast model from the autocomplete table |
| "All models rate-limited after N attempts" | All keys are at their limit | Increase debounceDelay and add keys from more providers |
Frequently asked questions
Is Continue free?
Yes. Continue is an open-source extension. It needs a model API. With FreeLLMAPI, it uses free provider tiers, so inference costs $0.
How do I set up free tab autocomplete in VS Code?
Install Continue and add a model with roles: [autocomplete], provider: openai, apiBase: http://localhost:3001/v1 and useLegacyCompletionsEndpoint: true. FreeLLMAPI answers on /v1/completions.
Which free model is best for Continue autocomplete?
The Continue docs recommend completion-trained models such as Codestral and Qwen Coder. Free offers for both are in the table on this page. Large chat and thinking models are not recommended.
Why does Continue not read my environment variable?
The Continue docs say that the IDE extensions cannot read shell environment variables. Put the key in ~/.continue/.env and refer to it as ${{ secrets.FREELLMAPI_API_KEY }}.
Does Continue agent mode work with free models?
Yes, if the model supports tool calling. Add the tool_use capability to the model in config.yaml and use a model with a large context window.
Related guides
- Free AI models in VS Code
- Cline, Roo Code and Kilo Code with a free API
- Best free coding model APIs
- Use free AI models in Zed
- Free OpenAI-compatible API base URLs
Sources
- Continue docs: OpenAI provider (checked 20 November 2026)
- Continue docs: config.yaml reference (checked 20 November 2026)
- Continue docs: autocomplete (checked 20 November 2026)
- Continue docs: secrets (FAQ) (checked 20 November 2026)
- FreeLLMAPI docs: VS Code ghost-text autocomplete (checked 20 November 2026)