Run an uncensored model in Zed’s Agent Panel

Add one openai_compatible provider to Zed’s settings.json, point it at the API, list the uncensored ids with their windows, store the key in the keychain or an env var. Then what a turn costs and how to read the errors.

Steps checked November 2, 2026 · prices as of November 16, 2026

Zed reads OpenAI-compatible providers from settings.json: an API URL, each model id with its context window, and a key kept in the keychain or an environment variable. Add one block and the Agent Panel runs an uncensored build with tool calls for file edits and terminal commands. The block in this guide already has the Unrestrictable URL and the ids Zed users call here most.

What you need

  • Zed with the Agent Panel, and write access to its user settings file. On macOS and Linux that is ~/.config/zed/settings.json; on Windows, %APPDATA%\Zed\settings.json. zed: open settings file in the command palette opens it. A project’s .zed/settings.json only accepts editor and language options, so providers always live in the user file.
  • A sk-unr- key from the console, with a balance behind it.
  • A model id and its context window from the models page. Zed cannot ask the API for either, so you type both.

Want to try the model before touching Zed? The same builds answer in the browser at /chat and in the Telegram bot, a few free turns a day with no account.

There is a settings UI as well. agent: open settings (or the menu at the Agent Panel’s top right) lands on Settings → AI → LLM Providers; Add Provider wants a provider name, an API URL, one model ID and its context window. That page also takes the key, which otherwise comes from an environment variable. The key is never written to settings.json.

Shortest path to a first reply

  1. Put the provider block in settings.json. Each entry under available_models must carry max_tokens, which is why the three ids here are the first builds in the list further down with a window the catalog publishes. Any other id from the models page fits the same shape, with the window its page shows. agent.default_model picks the first entry for new threads:
settings.json
1{
2 "language_models": {
3 "openai_compatible": {
4 "unrestrictable": {
5 "api_url": "https://api.unrestrictable.ai/v1",
6 "available_models": [
7 {
8 "name": "deepseek-v4.1-flash-1017-uncensored",
9 "display_name": "DeepSeek V4.1 Flash 1017 Uncensored",
10 "max_tokens": 1048576,
11 "capabilities": { "tools": true, "images": false, "parallel_tool_calls": false, "prompt_cache_key": false }
12 },
13 {
14 "name": "glm-5.3-flash-apex-abliterated",
15 "display_name": "GLM-5.3 Flash Apex Abliterated",
16 "max_tokens": 1048576,
17 "capabilities": { "tools": true, "images": false, "parallel_tool_calls": false, "prompt_cache_key": false }
18 },
19 {
20 "name": "kimi-k3-uncensored",
21 "display_name": "Kimi K3 Uncensored",
22 "max_tokens": 262144,
23 "max_output_tokens": 65536,
24 "capabilities": { "tools": true, "images": true, "parallel_tool_calls": false, "prompt_cache_key": false }
25 }
26 ]
27 }
28 }
29 },
30 "agent": {
31 "default_model": {
32 "provider": "unrestrictable",
33 "model": "deepseek-v4.1-flash-1017-uncensored"
34 }
35 }
36}
  1. Make a key in the console. Hand it to Zed either through the UI (agent: open settings, locate unrestrictable under LLM Providers, paste; it lands in the system keychain) or through an environment variable in the shell Zed starts from. Zed builds the variable name from the provider id:
shell
1export UNRESTRICTABLE_API_KEY="sk-unr-live-…"
  1. Open a thread (agent: new thread, or the sparkle in the status bar) and type something. The model selector sits on the message editor (cmd-alt-/ for macOS, ctrl-alt-/ for Linux and Windows) and swaps between the ids under unrestrictable, and whatever you pick is written back to agent.default_model.

When both exist, a non-empty environment variable beats the keychain. To drop a key that came in through the variable, clear the variable, then restart Zed.

The settings, one by one

openai_compatible.unrestrictable
the provider id
shows as the provider name in the selector, is the provider in agent.default_model, and seeds the env var name
api_url
https://api.unrestrictable.ai/v1
the request target; keep the trailing /v1 and add nothing after it
available_models[].name
deepseek-v4.1-flash-1017-uncensored
the id sent on every request, spelled as the models page spells it; the selector lists these ids and no others
available_models[].display_name
the name in the picker
optional; without it the picker shows the id
available_models[].max_tokens
the context window
required; by default Zed compacts a thread at 90% of this, so use the catalog figure and compaction fires at the right point
available_models[].max_output_tokens
the output limit
optional; goes out as max_completion_tokens on each request
available_models[].capabilities
{ tools, images, parallel_tool_calls, prompt_cache_key }
tools defaults to true and is required for edits and shell commands; images mirrors the catalog; the last two are left false
agent.default_model
{ provider: unrestrictable, model: deepseek-v4.1-flash-1017-uncensored }
the model a new thread opens on; inline_assistant_model, commit_message_model, thread_summary_model and compaction_model use the same shape
UNRESTRICTABLE_API_KEY
the whole sk-unr- key
the env var Zed checks for this provider id; the alternative is the keychain entry made in the settings UI

Which build to put first

Zed talks OpenAI-format chat completions to api_url; each build listed on the models page speaks that format. The bill for an agent turn is mostly input: the whole thread, every file Zed opened and every tool result ride along on each request, so a low input price is what makes a turn cheap. The right-hand column says whether the build accepts tool calls. Without them the selector marks the model No tools, and the agent can neither edit files nor run commands.

The window counts twice over. Compaction kicks in once a thread reaches 90% of max_tokens for its entry, the agent.auto_compact threshold. A long window means fewer compactions and more of the thread kept word for word. Under 80,000 tokens Zed does not auto-compact; it warns near the limit and offers a fresh thread seeded with a summary. Each compaction is one more request to the thread’s model, charged like the rest, unless agent.compaction_model names a smaller build for the job.

ModelOne agent turnTool calls
GLM-5.3 Uncensored$0.01model page
GLM-5.3 Flash Uncensored$0.0047model page
DeepSeek V4.1 Flash Abliterated$0.04model page
DeepSeek V4 Pro Uncensored$0.04model page
GLM-5.3 Abliterated$0.09model page
GLM-5.3 Flash Code Uncensored$0.25model page
DeepSeek V4.1 Flash 1017 Uncensored$0.04model page
GLM-5.3 Flash Apex Abliterated$0.04model page
Kimi K3 Uncensored$0.26model page

One turn here is 40,000 input tokens and 1,200 output tokens at catalog prices on 2026-11-16. Tool-call support is as the catalog states it; the model page covers the rest.

Uncensored models and their prices per million tokens
GLM-5.3 Uncensoredglm-5.3-uncensoredUncensored$0.282$0.846
GLM-5.3 Flash Uncensoredglm-5.3-flash-uncensoredUncensored$0.1053$0.3685
DeepSeek V4.1 Flash Abliterateddeepseek-v4.1-flash-abliteratedAbliterated$0.8488$1.6975
DeepSeek V4 Pro Uncensoreddeepseek-v4-pro-uncensoredUncensored$0.97$1.94
GLM-5.3 Abliteratedglm-5.3-abliteratedAbliterated$2.1825$4.365
GLM-5.3 Flash Code Uncensoredglm-5.3-flash-code-uncensoredUncensored$6.00$6.00
DeepSeek V4.1 Flash 1017 Uncensoreddeepseek-v4.1-flash-1017-uncensoredUncensored$0.8003$3.201
GLM-5.3 Flash Apex Abliteratedglm-5.3-flash-apex-abliteratedAbliterated$0.80$3.20
Kimi K3 Uncensoredkimi-k3-uncensoredUncensored$5.745$28.68

Prices per million tokens, read from the live catalog on 2026-11-16.

For a coding-specific comparison, the uncensored coding AI guide prices a full working day on each code build. These ids drop straight into Cursor and OpenCode, and reach Claude Code through a router.

Confirm it works

Ask the API which models your key may call. Each id under available_models should appear in the list:

shell
1curl https://api.unrestrictable.ai/v1/models -H "Authorization: Bearer sk-unr-live-…"

Now make one request shaped the way Zed makes them. The -i flag shows the response headers, which carry the request id and, on a non-streamed reply, the charge in USD.

shell
1curl -i https://api.unrestrictable.ai/v1/chat/completions \
2 -H "Authorization: Bearer sk-unr-live-…" \
3 -H "Content-Type: application/json" \
4 -d '{"model": "deepseek-v4.1-flash-1017-uncensored", "messages": [{"role": "user", "content": "Reply with the single word ok."}]}'

Back in Zed, start a new thread on one of your ids and send a message. The console’s usage page lists that request with its token counts and its charge. The panel’s message editor shows the thread’s running token count beside the profile selector, measured against the max_tokens you entered.

When it fails

  • 401 invalid_api_key: a wrong or revoked key, or a stale UNRESTRICTABLE_API_KEY in Zed’s environment that is shadowing the keychain entry. Paste a fresh copy from the console, or clear the variable before restarting Zed.
  • 404 model_not_found: one of the name values under available_models is missing from the models page. Compare the spelling; remember that name takes the id, while display_name takes the label.
  • 402 insufficient_balance: not enough balance for the request; top up from the console. 402 monthly_limit_reached is the monthly spend cap you put on the key.
  • 403 scope_denied: this key only covers certain models, and the one you asked for is outside that set.
  • 429 rate_limited: the key is over its rate limit. Wait the retry-after the response gives.
  • LLM provider is not configured or does not support the configured model: one of three things is off. agent.default_model (or inline_assistant_model) names a provider that differs from the key under openai_compatible; its model matches none of that provider’s available_models names; or no key is stored for the provider. Line up all three and add the key. Zed issue #63878, open since September 2026, describes the inline assistant showing this while the Agent Panel works on the identical provider and model.
  • No tools beside the model: that entry has capabilities.tools set to false. Switch it to true when the model page lists tool calls for the build.
  • The UI refuses to edit the provider: the LLM Providers page, per Zed issue #35461 (still open), can change a custom provider’s key and nothing else. URL, ids and windows are edited in settings.json.
  • Thread summaries or compaction fail with a reasoning_effort error: Zed puts "reasoning_effort": "none" on those requests whenever the entry declares a reasoning_effort (Zed issue #60411, open as of November 2026). Drop the field from such entries.
  • Two providers on one api_url fight over a single key: Zed keys the stored secret by URL, so a second block aimed at https://api.unrestrictable.ai/v1 with its own key prompts for it again and again (issue #41585). Use one block holding every id; restricting a key to some models is a scope set on the key, not a reason for a second provider.

Questions

Two places work. The LLM Providers page in Agent Settings puts it in the system keychain. Or export an environment variable named after the provider id in upper snake case with _API_KEY appended, which is UNRESTRICTABLE_API_KEY for the block in this guide. settings.json never holds the key, and a set variable overrides the keychain.

The provider and model named in agent.default_model (or inline_assistant_model) did not resolve to a configured provider that has a key. Check that provider matches the key under openai_compatible, that model is a name inside that provider’s available_models, and that a key exists for it in the keychain or the environment.

Yes. agent.inline_assistant_model, agent.commit_message_model, agent.compaction_model and agent.thread_summary_model each take { provider, model } like agent.default_model, and each falls back to the default when unset. Most people put a fast build on commits and summaries and a stronger one on the agent thread; the table above prices one agent turn per build. See the models page.

No. The selector shows exactly the ids under available_models, each with its max_tokens. When a new build lands on the models page, add its id and window to the block and it appears. The curl under Confirm it works prints every id the key can reach.

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.

Get a keyPrepaid from $10, no subscription.

New models and price changes, once a month.

New uncensored builds with their prices, price moves, and what the daily model check found.

Join on Telegram