To Continue, this API is one more models entry in config.yaml: provider: openai, a base URL, a key and a list of roles. Add that block once and an uncensored build answers in Chat, rewrites code inline and drives Agent mode across VS Code, JetBrains and the cn CLI. The same ids also run in Cursor, OpenCode and, behind a router, in Claude Code.
What you need
- Continue in VS Code or JetBrains, or the
cnCLI. All three read~/.continue/config.yaml. - A
sk-unr-key from the console, with a balance behind it. Keys are priced per token; see pricing. - A model id from the models page. The config below uses three.
No editor at hand? The same builds answer in the browser at /chat, with a few free turns a day and no account, and on Telegram.
The config file and how to reach it
Continue does not ship a provider called "OpenAI-compatible". Its docs treat any server that answers in the OpenAI format as provider: openai plus an apiBase that points elsewhere. This API fits that description. The file is ~/.continue/config.yaml on macOS and Linux, %USERPROFILE%\.continue\config.yaml on Windows, written on first launch.
From inside the editor: open the sidebar (Cmd/Ctrl + L in VS Code, Cmd/Ctrl + J in JetBrains), click the agent selector over the chat box, hover the agent and hit the gear. The configs dropdown at the top right of the chat box reaches the same file through the cog next to Local Config. Because cn reads this file too, you edit once for the IDE and the terminal.
The older config.json next to it is deprecated. Everything below is YAML, with one JSON example for a machine that has not moved yet.
Shortest path to a first reply
- Make a key in the console. Write it to
~/.continue/.envas a single line without quotes:
UNRESTRICTABLE_API_KEY=sk-unr-live-…That is where ${{ secrets.UNRESTRICTABLE_API_KEY }} gets its value. For a key tied to one project, use .env in the workspace root or .continue/.env below it. Continue looks there first, then in the global file, then in the process environment. Only the CLI sees that last source; the extensions cannot read shell variables.
- Paste these entries under
models:inconfig.yaml. Leave thename,versionandschemalines at the top alone. Any other id from the models page takes the same shape: its catalog display name onname, and acontextLengthwhen the catalog lists a window:
models: - name: GLM-5.3 Uncensored provider: openai model: glm-5.3-uncensored apiBase: https://api.unrestrictable.ai/v1 apiKey: ${{ secrets.UNRESTRICTABLE_API_KEY }} roles: [chat, edit, apply] capabilities: [tool_use] - name: GLM-5.3 Flash Uncensored provider: openai model: glm-5.3-flash-uncensored apiBase: https://api.unrestrictable.ai/v1 apiKey: ${{ secrets.UNRESTRICTABLE_API_KEY }} roles: [chat, edit, apply] capabilities: [tool_use] - name: DeepSeek V4.1 Flash Abliterated provider: openai model: deepseek-v4.1-flash-abliterated apiBase: https://api.unrestrictable.ai/v1 apiKey: ${{ secrets.UNRESTRICTABLE_API_KEY }} roles: [chat, edit, apply] capabilities: [tool_use]- Save the file. The entries show up in the model selector of the chat box. Choose one, then set Chat, Plan or Agent in the mode selector under the input;
Cmd/Ctrl + .steps through them. Edit and Apply pick up the same entries through their roles.
You can skip the secret and write the full sk-unr- key on the apiKey line; Continue documents both. For a file still on config.json, a single model reads as follows, with title in place of name and no roles:
{ "models": [ { "title": "GLM-5.3 Uncensored", "provider": "openai", "model": "glm-5.3-uncensored", "apiBase": "https://api.unrestrictable.ai/v1", "apiKey": "sk-unr-live-…" } ]}On the command line, cn loads the same config.yaml. Pass cn --config to use a different YAML for one run, or type /config in the TUI to swap between the local configs it has:
cn --config ~/.continue/config.yamlThe lines, one by one
- name
- the display namewhat Continue prints in the selector; no two entries may share one
- provider
- openaithe wire format; with it, Continue sends OpenAI chat completions to whatever apiBase you give
- model
- glm-5.3-uncensoredthe id in the request body, spelled as the models page has it
- apiBase
- https://api.unrestrictable.ai/v1the server that replaces OpenAI; stop at /v1, do not append /chat/completions
- apiKey
- ${{ secrets.UNRESTRICTABLE_API_KEY }}the full sk-unr- key, through a secret or pasted; goes out as the bearer token
- roles
- [chat, edit, apply]the places the entry is offered: chat and Agent, inline edit, writing a diff; autocomplete, embed and rerank are the others
- capabilities
- [tool_use]marks the model as taking tool calls, which Plan and Agent require; it extends what Continue detects, never removes it
- defaultCompletionOptions
- contextLengththe window Continue fits prompts into; maxTokens, temperature and stop live in the same map
- requestOptions
- timeout, caBundlePath, headersHTTP settings for this entry; a corporate certificate authority is handled here
Which build to put first
Every build on the models page speaks the OpenAI format, so each one works under provider: openai. What separates them in Continue is price per turn. By the time Agent has read a few files and run a few tools, one turn carries the whole conversation plus every tool result, so the input price decides the bill and a cheap input rate wins. The right-hand column says whether the build takes native tool calls; Continue's docs describe a system-message fallback for builds that do not, so capabilities: [tool_use] belongs on each entry regardless.
| Model | One Agent turn | Tool calls |
|---|---|---|
| GLM-5.3 Uncensored | $0.01 | model page |
| GLM-5.3 Flash Uncensored | $0.0047 | model page |
| DeepSeek V4.1 Flash Abliterated | $0.04 | model page |
| DeepSeek V4 Pro Uncensored | $0.04 | model page |
| GLM-5.3 Abliterated | $0.09 | model page |
| GLM-5.3 Flash Code Uncensored | $0.25 | model page |
| GLM-5.3 Uncensored | Uncensored | $0.282 | $0.846 |
| GLM-5.3 Flash Uncensored | Uncensored | $0.1053 | $0.3685 |
| DeepSeek V4.1 Flash Abliterated | Abliterated | $0.8488 | $1.6975 |
| DeepSeek V4 Pro Uncensored | Uncensored | $0.97 | $1.94 |
| GLM-5.3 Abliterated | Abliterated | $2.1825 | $4.365 |
| GLM-5.3 Flash Code Uncensored | Uncensored | $6.00 | $6.00 |
Each entry above carries chat, edit and apply: Chat, Plan and Agent, the inline edit, and the step that lands a proposed diff in the file. autocomplete is different. It has its own prompt template and request shape, and Continue's own example pairs a chat model with a separate autocomplete model. Keep these entries out of autocomplete and leave your current completion model in place. embed and rerank serve indexing and stay as they are too.
The uncensored coding AI guide prices a full day of coding on each code build. The ids here also drop into Cursor and OpenCode, each with its own settings screen.
Prove the key and the entry
Ask the API which models the key may call. Each id under models: has to appear in the list:
curl https://api.unrestrictable.ai/v1/models -H "Authorization: Bearer sk-unr-live-…"Next, send the request Continue would send. The -i flag prints the response headers, which carry an id for the request and, when the answer is not streamed, its cost in USD.
curl -i https://api.unrestrictable.ai/v1/chat/completions \ -H "Authorization: Bearer sk-unr-live-…" \ -H "Content-Type: application/json" \ -d '{"model": "glm-5.3-uncensored", "messages": [{"role": "user", "content": "Reply with the single word ok."}]}'Back in the editor, pick an entry, send one Chat message, then switch to Agent and request a small change. By default Agent stops and asks before each tool runs; tool policies in the docs can automate or block a tool. Plan works the same way with read-only tools, which makes it the place to let a new build walk the repository before it edits anything. Each request then appears on the console's usage page with tokens and charge.
When it fails
401 invalid_api_key: a bad or revoked key, or a secret that resolved to nothing. The extensions never see a variable exported in your shell; write it to~/.continue/.envunder the name in${{ secrets.UNRESTRICTABLE_API_KEY }}, or paste a fresh copy from the console.404 model_not_found: the value onmodelis not an id from the models page. Compare the spelling, and make sure the id sits onmodeland not onname.- A bare
404that names no model:apiBaselacks its/v1or carries an extra/chat/completions. Continue appends the route on its own, so the value ends at/v1. 402 insufficient_balance: not enough balance for this request; add funds in the console.402 monthly_limit_reached: the key reached the monthly cap you gave it.403 scope_denied: this key is scoped to other models.429 rate_limited: the key is over its rate limit;retry-afterin the response says how long to wait.- Plan or Agent shows Not Supported: Continue found no tool support on the entry. Per the docs,
capabilities: [tool_use]on the model fixes it, and the block above includes it. A build whose model page says no native tool calls then runs on Continue's system-message tools. - The entries are not in the selector: the configs dropdown is pointed at a different config. Set it to Local Config, which is
~/.continue/config.yaml. fetch failedbehind a company certificate authority: give each entryrequestOptions.caBundlePathwith the exported CA file, as Continue's FAQ describes.- Traffic lands on
/completionsrather than/chat/completions: the entry hasuseLegacyCompletionsEndpoint: true. Delete it; this API serves chat completions.
Questions
provider: openai together with apiBase set to the server’s /v1 URL. Continue ships no separate OpenAI-compatible provider; the docs send every server that speaks the OpenAI chat format through openai with an apiBase override, and this API is such a server.
On the apiKey line of the entry. Paste it whole, or write ${{ secrets.NAME }} and define NAME in ~/.continue/.env, in a workspace .env or in .continue/.env. Only the cn CLI reads shell variables; the VS Code and JetBrains extensions read the .env files.
No. This API speaks the OpenAI format under its own key, so Continue talks to it directly through the entry above. LiteLLM earns its place only when several providers need one local URL. See the LiteLLM guide.
Yes. Plan and Agent depend on tool calls, so every entry carries capabilities: [tool_use]. Builds whose model page lists native tool support take them directly; for the others, Continue’s docs describe a system-message fallback. The table above prices one Agent turn on each build from the live catalog. See the models page.
Yes. The same builds answer in the browser at /chat with a few free turns a day and no account, and the Telegram bot gives free replies as well. See open /chat.
Every model above is on the uncensored model list with its live price per million tokens. Per-token pricing sets the whole catalog side by side, and the family directory groups the builds by base model.