Continue with an uncensored model

Where the settings are

Continue has no separate OpenAI-compatible provider. Per its docs, a server that answers the OpenAI format is set up as provider: openai with an apiBase, and that is what this API is. The settings live in ~/.continue/config.yaml on macOS and Linux and %USERPROFILE%\.continue\config.yaml on Windows; Continue creates the file the first time it runs.

To open it from the IDE: open the Continue sidebar (Cmd/Ctrl + L in VS Code, Cmd/Ctrl + J in JetBrains), click the agent selector above the chat input, hover the agent and click the gear icon. The configs dropdown at the top right of the chat input has the same file behind the cog beside Local Config. The cn CLI reads the same file by default, so one edit covers the extension and the terminal.

config.json in the same folder is the older format and is deprecated in the docs; the YAML below is the current one, with the JSON shape shown once for a file that is still on it.

Steps

1. Create a key in the console and put it in ~/.continue/.env, one line, no quotes:

Continue resolves ${{ secrets.UNCENSOREDLLM_API_KEY }} from that file. A project-only key goes in .env at the workspace root or in .continue/.env under it; the docs check those two first, then the global file, then the process environment. The last one is for the CLI only: the VS Code and JetBrains extensions cannot read shell variables.

2. Add these entries under models: in config.yaml, keeping the file’s name, version and schema lines as they are. The three ids are the ones Continue users call most here; any id on the models page goes in the same way, with its display name from the catalog and its context length where the catalog publishes one:

3. Save. The three names appear in the model selector of the chat input; pick one, then choose Chat, Plan or Agent in the mode selector below the input (Cmd/Ctrl + . cycles). Edit and Apply use the same entries through their roles.

To paste the key instead of using a secret, put the whole sk-ullm- key on the apiKey line; the docs show both forms. The same entry in the deprecated config.json, for one model, looks like this (title instead of name, no roles):

In the terminal, cn starts with the same config.yaml; cn --config points it at another YAML file for one session and /config inside the TUI switches between the local configs it knows:

What each setting does

name

the display name

the label Continue shows for the entry; it has to be unique in the file

provider

openai

the request format; Continue speaks OpenAI chat completions to any apiBase under it

model

glm-5.3-uncensored

the id sent in each request, exactly as the models page spells it

apiBase

https://api.uncensoredllm.ai/v1

where the requests go instead of OpenAI; ends in /v1, never in /chat/completions

apiKey

${{ secrets.UNCENSOREDLLM_API_KEY }}

the whole sk-ullm- key, by secret name or pasted; sent as the bearer token

roles

[chat, edit, apply]

where the entry can be picked: chat and Agent, inline edits, applying a diff; autocomplete, embed and rerank are the other roles

capabilities

[tool_use]

tells Continue the model takes tool calls, which Plan and Agent need; it only adds to what Continue detects on its own

defaultCompletionOptions

contextLength

the window Continue budgets prompts against; maxTokens, temperature and stop go in the same map

requestOptions

timeout, caBundlePath, headers

per-model HTTP settings; the fix for a custom certificate authority lives here

Picking a model

Continue sends OpenAI-format requests under provider: openai, and every build on the models page answers in that format. An Agent turn carries the conversation, the files Continue read and every tool result so far, so input tokens are most of the bill and a build with a low input price costs the least per turn. The last column says whether the build takes native tool calls. Continue’s docs say it falls back to system-message tools for a model without them, so capabilities: [tool_use] stays on every entry either way.

40,000 input and 1,200 output tokens at the catalog prices on 2026-11-09. Tool calls as the catalog publishes them; where it does not, the model page says.

The block gives each entry the chat, edit and apply roles, which cover Chat, Plan and Agent, the inline edit and the step that writes a proposed diff into the file. autocomplete is a separate role with its own prompt template and its own request shape; the docs’ own example pairs a chat model with a dedicated autocomplete model, and that split holds here. Leave these entries off autocomplete and keep whatever fills that role today. embed and rerank are indexing roles and stay where they are as well.

For coding specifically, the uncensored coding AI guide compares the code builds and costs a full coding day on each. The same ids work in Cursor and OpenCode with their own settings.

Check it

List the models your key can call; every id you put under models: should be in the answer:

Then send one request the way Continue does. -i prints the headers: x-uplink-request-id names the request, and on a non-streamed answer x-uplink-charge is what it cost in USD.

In Continue, send a message in Chat with one of your entries selected, then switch to Agent and ask for a small edit. Agent asks before it runs a tool by default; the docs’ tool policies make a tool automatic or exclude it. Plan does the same with read-only tools, so it is the safe place to watch a new build read the repository before it changes anything. The requests show on the console’s usage page with their tokens and their charge.

Errors and fixes

• 401 invalid_api_key: the key is wrong or revoked, or the secret did not resolve. A key exported in the shell is invisible to the extensions; put it in ~/.continue/.env under the name used in ${{ secrets.UNCENSOREDLLM_API_KEY }}, or copy it again from the console.
• 404 model_not_found: the model line is not an id on the models page. Check the spelling against the page; the id goes in model, not in name.
• 404 with no model in the message: apiBase is missing its /v1, or has /chat/completions appended. Continue adds the route itself; apiBase ends at /v1.
• 402 insufficient_balance: the balance does not cover the request. Top up in the console. 402 monthly_limit_reached means the key hit the monthly spend limit you set on it.
• 403 scope_denied: the key is limited to certain models and this is not one of them.
• 429 rate_limited: the key hit its rate limit. The response carries retry-after.
• Plan or Agent greyed out with Not Supported: Continue did not detect tool support on the entry. The docs’ fix is capabilities: [tool_use] on the model, which the block above already has; a build whose model page says it takes no tool calls runs on Continue’s system-message tools instead.
• The entries are missing from the model selector: the configs dropdown is on another config. Switch it to Local Config, the one backed by ~/.continue/config.yaml.
• fetch failed on a network with its own certificate authority: set requestOptions.caBundlePath on each entry to the exported CA file, per the FAQ in Continue’s docs.
• Requests arriving at /completions instead of /chat/completions: the entry carries useLegacyCompletionsEndpoint: true. Remove it; the API serves chat completions.

Questions