The Neconyan handbook
Your guides Miso, Taro & Nori
Choose each guide's gender to change their artwork and pronouns. Your choices stay in this browser, separately from the app.
Connect a model¶
Neconyan sends your conversation to a model and displays the reply. You choose the service, model and credentials through Connections in the workspace sidebar.
Choose the kind of connection¶
| Connection | What you need | Where the writing runs |
|---|---|---|
| Online provider | An account with API access, any required credit and an API key | The provider's servers |
| Local model service | A running model service and its supported connection address | Your own hardware |
| Compatible custom endpoint | Its API format, address, model name and any required key | Wherever that endpoint is hosted |
An endpoint is the address of a service Neconyan can send requests to. 'OpenAI-compatible' describes an API format; it doesn't mean the service is operated by OpenAI or accepts an OpenAI key.
Make the first connection¶
- Open Connections. On a phone, open the workspace navigation first.
- Select the API type and provider that match the service you're using.
- Enter its API key when required. Copy the key from that provider's account page, without extra spaces.
- Choose an available model from the searchable list, or enter its exact identifier if the connection requires it.
- If you're using a custom or local endpoint, enter the address in the format that connection expects. Follow its example rather than adding an extra path by habit.
- Connect, then try a short message with a bundled character.
A model list loading successfully checks only part of the connection. A real reply confirms that the selected model accepts a generation request with your current settings.
Miso
Try a tiny request first: 'Give me one cosy scene idea.' Once that works, we can add your lorebook.
Find a model in a long list¶
Every chat and text-completion connection with a single model choice shows its models as a list of buttons, with a Search by name or model ID… box above it. The list works the same on a phone and a computer.
Type any part of a name or ID to narrow it. Every word you type has to match, and case doesn't matter. Escape clears the search, Enter chooses the first match when the box has text, and the arrow keys move through the rows. Choosing a row selects that model straight away.
Models marked as favourites move into a Favourites group with a star at the right edge. The rest are grouped by family worked out from their IDs, so a new version such as gemini-3.5-flash joins Gemini 3.5 without waiting for an update. OpenRouter and ElectronHub can keep their own vendor groups. NanoGPT labels start with the source, such as Google/Gemma 4 31B, because the same model can come from more than one organisation.
The list shows every model the provider returns; scroll to browse or search to narrow it. Searching only changes the view. It saves nothing and makes no model request.
Save a connection profile¶
A connection profile remembers a model setup so you can return to it or assign it to a helper. Give it a useful name that includes the provider and purpose, such as 'Local writing' or 'Quick summaries'.
The main chat and background helpers can use different connections. Scratchpad, Agents and Mewmory each have their own connection choices. Check those separately if the main chat works but a helper fails.
Switching profiles can also change the model and related request settings. Read the active connection before sending a long or expensive request.
Confirm a proxy¶
A connection profile or Chat Completion preset that carries a reverse proxy asks Connecting To Proxy when you create or update it, or when you import a preset with one. Choose Yes to save it, or No to stop: the change isn't saved, and a preset isn't saved or imported, with a message saying why.
Once you confirm an address for this account, Neconyan doesn't ask again. A preset saves the exact address you approved, even if the field changes while the question is open. A profile saved before this change asks once on its first request.
Using a local model¶
Start the model service before connecting Neconyan. Confirm which API formats it offers, the port it's listening on and which model is loaded. Neconyan is the chat interface; your model service manages the actual model files and hardware.
The endpoint must be reachable from the machine running the Neconyan server. If the server is on your computer and you're visiting from a phone, the endpoint is still interpreted from the computer. If Neconyan itself runs on Android, a localhost address refers to the phone.
Provider notes¶
OpenAI¶
Choose Chat Completion, then OpenAI under Chat Completion Source. Create a key on the API keys page of your OpenAI platform account; a ChatGPT subscription doesn't include API access or credit. Paste the key into OpenAI API key, connect, then pick a model from Available Models or type its exact Model ID.
View API Usage Metrics opens your usage on OpenAI's site. If you're connecting through a proxy, use its Proxy Password field; the key box is ignored.
DreamGen¶
Choose Text Completion, then DreamGen as the API Type. Create an API key in your DreamGen account and paste it into DreamGen API key. The DreamGen Model list reads -- Connect to the API -- until the connection succeeds, then lists the models your account can use.
Neconyan doesn't include a DreamGen template. Check DreamGen's documentation for the prompt format its model expects and match it in Formatting.
NovelAI and AI Horde¶
Neconyan no longer offers NovelAI, AI Horde or KoboldAI Classic as reply connections. If saved settings still name one, Neconyan switches to Chat Completion; choose another provider there or under Text Completion. NovelAI is still available for pictures in Quick Image Gen.
AI Horde ran requests on computers lent by volunteers, who could see what was sent. Bear that in mind with any shared or community-run service: don't send anything you wouldn't want a stranger to read.
Where keys are kept¶
After you connect, the key box empties and shows Key saved, or Missing key if there's none. Keys are stored by the Neconyan server for your account, not in the page. The key button beside each box opens Manage API keys, where you can keep several labelled keys for one provider and choose which is used.
View hidden API keys at the bottom of Connections only works if the server owner has allowed it in config.yaml. Data imports never bring keys or passwords with them.
Understand the limits¶
Context capacity is how much input and output the model can handle together. Reply limit is the maximum output you ask it to produce. A larger reply limit doesn't enlarge its context capacity and can leave less room for history.
Start with settings the provider supports. Some models reject particular sampling options, image inputs, tools or requested output sizes. If you see a request error, use its exact wording to identify the unsupported setting rather than changing everything at once.
Costs and privacy¶
The connected service receives the material included in the request: relevant chat history, character information and any enabled context. Helpers and image generation may make additional requests. Check your provider's pricing and data policy, and use context previews where a feature offers them.
Keys are credentials. Don't paste them into a character card, public screenshot or bug report. For errors, share the provider type, model identifier and redacted error message instead.
Next: Your first chat. If a request is failing, see Something went wrong.