Aider is the pair programmer that lives in your terminal and edits the repo directly. It speaks OpenAI's chat format, so an uncensored build from this catalog needs a base URL, a key and the openai/ prefix on the model id, nothing more. Below: the two-command start, the config files, the cost readout and the errors. The ids carry over to OpenCode, Cursor and, via a router, Claude Code.
What you need
- An API key from the console. It begins with
sk-unr-and draws on a prepaid balance, per token. - Aider installed, and a project folder to open it in.
- A model id copied from the models page. The examples below use
glm-5.3-uncensored. - To try a build before installing anything, chat with it at /chat, free each day with no account, or in Telegram through the bot.
Two commands to a first edit
Aider has no provider picker. It reads an OpenAI base URL and key from the environment, and treats any model written as openai/<id> as living at that URL. Export both values in the shell you will launch aider from (Windows: setx, then open a fresh shell):
export OPENAI_API_BASE=https://api.unrestrictable.ai/v1export OPENAI_API_KEY=sk-unr-live-…Then open your project and name the build behind the prefix:
cd /to/your/projectaider --model openai/glm-5.3-uncensoredAsk for a change. Aider sends the request to this API, receives search/replace blocks and writes them into your files. Swap the id for any other on the models page and nothing else moves.
Where aider looks for settings
Every flag can live in .aider.conf.yml as a YAML key. Aider reads that file from three places: your home directory, the root of the git repo and the folder you launched from, in that order, with a later file overriding an earlier one. --config <file> reads a single file and ignores the rest.
Environment variables have the same three homes in a .env file, or one file of your choosing through --env-file. Each flag also maps to an AIDER_ variable, so AIDER_MODEL and AIDER_OPENAI_API_BASE can sit in a shell profile.
For a model aider has never heard of, two more files fill in what it would otherwise guess. .aider.model.metadata.json gives the context window and the prices; .aider.model.settings.yml gives the edit format, the weak model and the repo map. Both are searched in the same three places, or wherever --model-metadata-file and --model-settings-file point. Mid-session, /model, /weak-model and /editor-model change the builds, and /settings shows what aider is running with.
Move the flags into files
- Write the base URL, key and models into
.aider.conf.ymlat the repo root, and the command shrinks toaider. Theweak-modelline names the build for commit messages and history summaries, work a fast build does well for less. If the key already sits in the environment, drop theopenai-api-keyline:
# .aider.conf.yml, in the git root (or ~ for every project)openai-api-base: https://api.unrestrictable.ai/v1openai-api-key: sk-unr-live-…model: openai/glm-5.3-uncensoredweak-model: openai/glm-5.3-flash-uncensored- On launch aider warns
Unknown context window size and costs, using sane defaultsfor an id outside its own table. Edits still work; what you lose is the token budget and the running cost. Supply both in.aider.model.metadata.jsonbeside the config, each entry keyed with theopenai/prefix. The numbers here come straight from the catalog for the three ids, fetched when this page was built:
{ "openai/glm-5.3-uncensored": { "input_cost_per_token": 0.000000282, "output_cost_per_token": 0.000000846, "litellm_provider": "openai", "mode": "chat" }, "openai/glm-5.3-flash-uncensored": { "input_cost_per_token": 0.00000010528, "output_cost_per_token": 0.00000036848, "litellm_provider": "openai", "mode": "chat" }, "openai/deepseek-v4.1-flash-abliterated": { "input_cost_per_token": 0.00000084875, "output_cost_per_token": 0.0000016975, "litellm_provider": "openai", "mode": "chat" }}- Optional.
.aider.model.settings.ymlpins the edit format and the helper builds to each id, so they survive a/modelswitch.diffmeans search/replace blocks: only the lines that change come back:
- name: openai/glm-5.3-uncensored edit_format: diff weak_model_name: openai/glm-5.3-flash-uncensored editor_model_name: openai/deepseek-v4.1-flash-abliterated editor_edit_format: editor-diff use_repo_map: true- name: openai/glm-5.3-flash-uncensored edit_format: diff use_repo_map: true- name: openai/deepseek-v4.1-flash-abliterated edit_format: diff use_repo_map: true- Optional. Architect mode splits each turn in two: the main build plans the change, then the editor build turns the plan into edit blocks. Twice the requests, so a higher bill, but it rescues a build that reasons well and botches the edit syntax. In the chat,
/architectapplies the same split to a single message:
aider --architect --model openai/glm-5.3-uncensored --editor-model openai/deepseek-v4.1-flash-abliteratedSetting by setting
- OPENAI_API_BASE
- https://api.unrestrictable.ai/v1every request goes here, and the trailing /v1 is required. In YAML: openai-api-base; on the command line: --openai-api-base
- OPENAI_API_KEY
- the whole sk-unr- keysent as the bearer token on each call. YAML openai-api-key, flag --openai-api-key. Aider treats it as its OpenAI key, so --api-key is the wrong slot
- --model
- openai/glm-5.3-uncensoredthe build that writes the code. openai/ routes it to the base URL; what follows must match the models page letter for letter
- --weak-model
- openai/glm-5.3-flash-uncensoredhandles commit messages and history summaries. Aider normally derives it from the main model, which it cannot do for an unknown id
- --architect / --editor-model
- openai/deepseek-v4.1-flash-abliteratedthe planner and the editor of a two-step turn; --editor-edit-format chooses editor-diff or editor-whole for the second step
- .aider.model.metadata.json
- openai/<id> → window and costper openai/<id>: max_input_tokens, max_output_tokens and the two per-token costs. Silences the unknown-model line and switches the cost counter on
- .aider.model.settings.yml
- openai/<id> → edit_format…per openai/<id>: edit_format, weak_model_name, editor_model_name, use_repo_map and the other ModelSettings fields aider lacks for an id it does not ship
- --show-model-warnings
- true by defaulton unless you pass --no-show-model-warnings, which hides the line but leaves the cost counter off
Which build to run
Every build on the models page speaks the chat format aider sends. What a turn carries: the system prompt, the repo map (about --map-tokens worth, 1k unless you change it, larger while no files are in the chat), whatever you pulled in with /add, and the whole conversation so far. What comes back is a handful of search/replace blocks. The bill is mostly input tokens, so the input price decides the cost of a turn, and /drop or /clear shrinks the next one at once.
Because diff and whole are plain-text formats, a build without tool calling still edits files fine. The tool-call column is for later, when you reuse the id in OpenCode or Cursor, whose agents cannot run without it.
| Model | Cost of one 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 |
The usual pairing is a strong build under --model and a fast one under --weak-model; with the metadata file in place aider totals the session as you go. For a wider comparison of the code builds, and what a full day of coding costs on each, read the uncensored coding AI guide.
Prove the wiring
First ask the API which models your key can reach. Every id you plan to put behind openai/ must appear:
curl https://api.unrestrictable.ai/v1/models -H "Authorization: Bearer sk-unr-live-…"Next, one chat request shaped like aider's own. With -i curl prints the response headers, which carry a request id and, when the answer is not streamed, the charge 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."}]}'Inside aider, the opening lines announce the main build, its edit format and the weak build. /settings dumps the whole configuration; /tokens breaks the next request down file by file. Add one file, send one message, and the call appears on the console's usage page with its token counts and charge.
When something fails
401 invalid_api_key: the key is mistyped or revoked, or the shell that launched aider never sawOPENAI_API_KEY. Export it once more, or paste a fresh copy from the console. In YAML the key belongs underopenai-api-key:; anapi-key:line lands somewhere else.404 model_not_found: whatever followsopenai/is not an id on the models page. Compare it character by character; everything after the prefix is the id.402 insufficient_balance: not enough prepaid balance for this request. Top up in the console.402 monthly_limit_reachedis different: the key reached the monthly cap you gave it.403 scope_denied: this key is restricted to a list of models, and the one you asked for is not on it.429 rate_limited: too many requests for this key. Wait theretry-afterthe response names.Unknown context window size and costs, using sane defaults: aider lacks an entry for the id. Nothing is broken; it proceeds as if the window were unlimited and the tokens free. Step 2 of the files section fixes it with real figures, or--no-show-model-warningshides the message.- Aider offers a model with a similar name and will not start with yours: you left off
openai/. The prefix is the routing; a bare name is looked up in aider's built-in list instead. Failed to apply edit to <file>: the build broke the edit format. Aider's own ladder: a stronger build, then--edit-format whole, then architect mode with a separate editor build, then fewer files in the chat (/drop,/clear, with/tokensto measure). Aider's docs say most models begin to slip past roughly 25k tokens of context..aider.model.settings.ymlseems ignored: eachnamemust carry the prefix,openai/glm-5.3-uncensored, never the id on its own.
Questions
Yes. Point OPENAI_API_BASE at the endpoint, put the key in OPENAI_API_KEY, and write the model as openai/<id>. Aider routes every prefixed id to that base URL, which makes this API a drop-in; the YAML keys hold the same values when you prefer a file.
The id is not in aider’s built-in table. Give it one in .aider.model.metadata.json, keyed openai/<id>, with the window, the output cap and the two per-token prices. The warning stops and the running cost appears. --no-show-model-warnings only hides the line.
diff first: search/replace blocks that carry only the changed lines. When a build keeps failing to apply them, try --edit-format whole, which rewrites the entire file at a higher output cost, or --architect, which hands the edit blocks to a second build.
Whatever the tokens come to at the model’s listed price, drawn from your prepaid balance; there is no subscription. The turn table above prices one edit-sized exchange per build at today’s rate, and the weak build bills commit messages and summaries at its own lower price. See the models page.
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.