With Proxy selected, a HammerAI character talks to a model that lives outside HammerAI. This guide puts an uncensored roleplay build behind it: four fields, a URL, an id, a context length and a key, and the chat is running with nothing installed and no GPU on your machine. Field names were read from HammerAI's Proxy Mode doc on 24 October 2026.
New to builds with the refusals removed? Read AI with no restrictions first, then connect one here.
What you need
- HammerAI, in the browser or as the desktop app. Proxy is available on the Free plan: HammerAI's pricing page shows "Proxy to any LLM API" on Free and on each paid tier (read 24 October 2026).
- A key from the console. It begins
sk-unr-and is displayed a single time, so store it before closing the dialog. Top up from $10 by card, WeChat Pay or crypto. A key with no balance is accepted by the form, but its first message comes back 402. - A model id, copied from the uncensored model list or the table further down. Copy, do not type: Proxy Model has no dropdown and checks nothing.
Want to try a build before wiring up HammerAI? The browser chat at /chat gives three free turns a day with no account, with characters built in, and the Telegram bot gives three free replies. The roleplay page has more on that side of the catalog.
Connect in nine steps
- Make the key in the console and copy it.
- Choose an id, for example
cydonia-24b-v4.3. - In HammerAI, open a character chat and click the Settings tab in the panel on the right.
- Open the LLM dropdown and scroll to its last entry, Proxy. Four new fields show up.
- Proxy URL:
https://api.unrestrictable.ai/v1. - Proxy Model: the id, pasted as is.
- Proxy Context Length: the build's window from the table below. Where the catalog publishes none, leave the field empty, which is what HammerAI's doc says to do for an unknown window.
- Proxy API Key: the key.
- Send a line. If no reply arrives, HammerAI's doc says to check endpoint, model name and key in turn, and adds that some services only connect after a full reload of the app.
That chat now sends each prompt to this API and shows the completion it gets back. Nothing downloads, and your own GPU is idle: the model runs here.
The four fields in one table
The labels are HammerAI's, taken from its Proxy Mode doc on 24 October 2026. The values are the ones for this API.
- Proxy URL
- https://api.unrestrictable.ai/v1Base URL; stops at /v1
- Proxy Model
- cydonia-24b-v4.3An id from the models page, character for character
- Proxy Context Length
- 131072Optional; the window in the table below
- Proxy API Key
- sk-unr-live-…From the console; displayed a single time
Note the URL shape. Every example in HammerAI's doc, for any service, stops at /v1; none runs on to /chat/completions. Follow that here.
Where HammerAI Proxy fits
HammerAI is a character and story chat app, on the web and as a desktop build for macOS, Windows and Linux. Its docs call it a platform for roleplay, image generation and interactive stories on top of language models. You can run a model locally in the desktop app, or use the cloud models HammerAI hosts, with more of them available on paid plans.
Proxy is the third route. Each prompt goes to an OpenAI-compatible API of your choosing, so the model, its window and its price belong to that API rather than to HammerAI. That matters for context: HammerAI's pricing page limits its own cloud models to 8,192 tokens on Free and 65,536 on the top plan, while a proxied model is limited only by its own window.
Four ids to try
Each build below is tagged roleplay in the catalog and speaks the OpenAI wire format HammerAI's Proxy expects. Prices and windows are pulled from the live catalog as the page renders.
| Wire formats | |||||
|---|---|---|---|---|---|
| Cydonia 24B v4.3 | Roleplay | $0.484 | $0.60 | 131K | |
| Magidonia 24B v4.3 | Roleplay | $0.3535 | $0.4235 | 33K | |
| Gemma 4 31B Fabled | Roleplay | $0.35 | $1.575 | 262K |
| Model id | Context window in the catalog | Proxy Context Length to enter |
|---|---|---|
cydonia-24b-v4.3 | 131,072 tokens | 131072 |
magidonia-24b-v4.3 | 32,768 tokens | 32768 |
gemma-4-31b-fabled | 262,144 tokens | 262144 |
gemma-4-31b-fabled has siblings; the uncensored Gemma page lists the other Gemma 4 tunes. Treat the window as a limit rather than a goal. HammerAI fills the context length you set with as much chat history as fits, and all of it is billed as input on every message, so a value below the window makes a long chat cheaper.
The cost of a night of chat
Take 200 messages in one session with replies near 250 tokens. The first column sends about 4,000 tokens of character card and history with each message. The second sends 16,000, which is where a lorebook and a long thread land.
| Model | 200 messages at 4,000 context | 200 messages at 16,000 context |
|---|---|---|
| Cydonia 24B v4.3 | $0.42 | $1.58 |
| Magidonia 24B v4.3 | $0.30 | $1.15 |
| Gemma 4 31B Fabled | $0.36 | $1.20 |
Nearly all of it is input. 200 × 4,000 comes to 800,000 input tokens, next to 50,000 of output. Trimming the character card and lowering Proxy Context Length change the total more than switching models does. Rates for every build are on the pricing page.
Test the key and the id first
One free request from a terminal lists the ids your key may call. Put your key in and run it before you touch HammerAI.
curl https://api.unrestrictable.ai/v1/models \ -H "Authorization: Bearer sk-unr-live-…"A 200 with ids in it clears the key. If your Proxy Model id appears in that list, the URL is the only field that can still be wrong. A 401 points at the key, and the body says how:
HTTP 401{"error":{"message":"No API key was sent.","type":"authentication_error","param":null,"code":"invalid_api_key"}}HTTP 401{"error":{"message":"The API key is not valid.","type":"authentication_error","param":null,"code":"invalid_api_key"}}When HammerAI Proxy returns nothing
401, a key problem
"No API key was sent." says Proxy API Key is blank. "The API key is not valid." says the key is mistyped, truncated or revoked. Make a new one in the console and paste the whole string.
404 model_not_found, an id problem
Proxy Model holds something the catalog does not: a typo, a display name where an id belongs, or a build that has been withdrawn. Run the terminal request and copy an id from its output.
402, a balance problem
The prepaid balance has run out. Top up in the console. A 402 reply is never billed.
CORS errors in the browser
In the web app the browser itself calls the API. This API accepts requests from any origin and allows the Authorization header (verified 24 October 2026), so no setting needs changing. The CORS note in HammerAI's doc concerns a local LM Studio server only and does not apply to this connection.
Every day the daily real-model check sends one fixed prompt to each model, so the build behind an id is the build on its label. The docs page carries the base URL, key and model id for other clients.
Questions
No. Its pricing page shows "Proxy to any LLM API" on its Free plan and on each paid plan (read 24 October 2026). What you pay is the API behind the Proxy, here billed per token on a prepaid balance. See per-token prices.
No. In Proxy mode HammerAI sends the prompt to the API and the model runs on that side. A GPU matters only for the local models in HammerAI's desktop app.
https://api.unrestrictable.ai/v1, the base URL ending in /v1. Leave /chat/completions off; every example in HammerAI’s doc is a base URL.
The window the catalog lists for the model, or something smaller to keep a long chat cheap. If the catalog lists no window, leave it blank, which is HammerAI's own advice.
Any id in the catalog works, since all of them take the OpenAI wire format. For character chat, begin with one of the roleplay-tagged builds; the table above gives four with windows and prices. See the uncensored model list.
These builds are abliterated or fine-tuned so that refusals are gone from the weights themselves. No prompt trick is involved, and there is nothing to paste into the character card. See AI with no restrictions, explained.
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.