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.
Included tools¶
Find documented macros in the macro reference.
Neconyan ships with a collection of writing, appearance and diagnostic tools. They sit alongside the rest of the workspace in the Workspace, Fine-tuning, Characters & Lore, Appearance and Troubleshooting sidebar sections. Open Extensions → Built-in for their settings pages, or Extensions → Manage extensions to inspect versions, authors, licences and enabled states. If a bundled tool is switched off, its sidebar button shows Off and opens Manage extensions so you can turn it on.
Included code updates with Neconyan. Third-party extensions you install separately have their own update path. A bundled tool can still use a paid model or an external service.
Writing and prompt work¶
| Tool | Use it for |
|---|---|
| Guided Generations | Give a response, swipe, regeneration, correction or impersonation specific guidance |
| Input History | Search earlier composer input and pick one to replace your draft |
| Quick Replies | Buttons above the message box for saved messages, steering and commands |
| Dice Roller | Roll d4 to d100 with up to four dice, a bonus, difficulty and advantage or disadvantage, then send the result to the story |
| Deep Swipe | Work with message alternatives, including supported user-message alternatives |
| Preset Tools | Search and organise prompt sections |
| Chat Completion Tabs | Switch separate Parameters and Prompts tabs on and off from the sidebar, or open its settings in Extensions → Built-in |
| Prompt Tags | Manage tags and defaults for supported prompt sections |
| Macro Enhanced | A full page for macros: try them in Playground, look them up in Reference, save your own in My macros, check this chat's saved values in This chat and set pronouns in Setup. See Macros |
| Prompting Lab | Build prompt tests, compare requests and run selected model comparisons |
Prompt reconstruction and dry diagnostics don't generate replies. Prompting Lab's model comparisons and optional analysis make real requests. Read the action before running a batch.
On phones, Prompting Lab and World Info Lab use visible section buttons instead of a section dropdown. Tap a section to open its controls without opening the keyboard.
Quick Replies¶
Quick Replies is the bundled extension; its settings header says Quick Reply. Buttons sit above the message box, and Show quick reply buttons turns the bar on. Sets are listed under Shown in every chat, Only in this chat and Only with this character; a character association is kept in your settings, not in the character card.
For a single button, use Make a quick reply. For a prepared set, use Ready-made buttons. Edit a set chooses the set you are editing, Set options holds its shared behaviour and Add quick reply adds another button to it.
The ready-made sets are Story controls (Continue, Try again, Write for me), Reply directions (such as More detail, Shorter and Time skip), Writing helpers (Action, Speech, OOC) and Ask privately (Recap, Ideas, Who is here?). Add set creates a set under a unique name, switches the bar on, and links the set to every chat when it isn't linked anywhere yet. Adding a set makes no model request; buttons that generate or ask do.
The builder asks for a Button label, What should it do? and the text, files the button under Add it to and saves it with Add button. The modes are Send it as my message, Put it in the message box, Steer the next reply, Ask the AI privately and Run a command. The chips insert What I typed, Character name and My name.
Steer the next reply sends anything already in the message box first, then asks for a reply that follows a one-use note. The note isn't added to the chat. Ask the AI privately still makes a model request: the answer appears in a pop-up instead of the chat, and the provider works as usual.
The builder refuses a button that doesn't fit the selected set. A set with Put replies in the message box instead of sending or Add what I typed automatically only accepts send-type buttons; the notice names the set and your options, and the set's own options are left alone.
Dice Roller¶
Dice Roller is on by default, and its dice button sits first beside the message box. Tap it to open the dice and roll straight away unless Roll as soon as the dice open is off; Roll dice is also in the wand menu. Type what you're attempting, then pick a Die from d4 to d100, up to four Dice, a Bonus from -20 to +20, a Difficulty (Easy, Normal, Hard, Extreme or Custom) and Luck: Advantage keeps the higher of two rolls, Disadvantage the lower. Add to message puts a line such as 🎲 Pick the lock - rolled 17 vs 12 (1d20+2: 15) - SUCCESS in the message box, and Send sends it. Replies get a short system note telling the model to write that outcome as settled fact without mentioning dice or numbers; continuing or swiping keeps it. It works with browser and server replies. Switch it off in Extensions → Manage extensions under Built-in features, or adjust the dice button, Roll as soon as the dice open, critical rolls and partial successes in Extensions → Dice Roller.
Lore and saved work¶
Pawthfinder chooses relevant lore with a separate decision model. In its settings, choose a Jev / decision model preset, then connect its endpoint or a saved profile. Jev scores which entries to retrieve; summary writing still uses your chat model. Decision protocol must match the provider's decision endpoint or chat-completion service.
If Pawthfinder should choose the lore by itself, enable Disable World Info keyword activation. Otherwise ordinary keyword matches can add entries too. Start with the preset's threshold and change one setting at a time.
Reuse the lore when swiping is on by default and needs Tool Mode or Predictive Pipeline. A swipe or regenerate then keeps the lore Pawthfinder already picked for the reply you're changing, instead of paying its decision model to choose that lore again. Turn it off to pick lore fresh for every swipe. A changed message, a changed lorebook, a different chat or a failed retrieval still picks lore again, and the reused answer lives in the open page, so the first swipe after a reload asks again.
World Info Lab traces lore activation and supports reviewed batch changes. It and Prompting Lab can open saved chats up to 64 MiB. Lorebook Distiller proposes lore entries from a saved chat. Chat Archive searches and organises chats; its organisation export doesn't contain the chat contents themselves.
Card & Lorebook Time Machine keeps supported snapshots of characters, lorebooks and presets. It isn't a whole-account or chat backup. Its window is a list and detail view now: snapshot cards, kind filters, a history timeline and field-by-field comparisons. On a phone it swaps between the list and one snapshot instead of squeezing both. Compare before restoring, because a restore replaces newer edits.
You can keep chatting and save settings while Snapshot everything now runs. Its snapshot update keeps your newer local choices and doesn't count as a settings change from another device. A real change from another device still stops a conflicting save.
If the item changes again after comparison, the restore stops rather than overwriting the intervening edit. Compare the current item again before retrying. A snapshot that fails its integrity check can't be used as a trusted restore source.
Several tools keep accepted work and results on the server. Closing the page stops watching it; use the explicit Stop action when you want to request cancellation.
Appearance and media¶
Dialogue Colors assigns speaker colours. Regex Agent Themes restyles trackers and Companion panels. CSS Snippets manages your small visual customisations. Termeownal UI offers an optional terminal-style interface with its own command glossary.
Quick Image Gen needs its own provider configuration. Its panel is built around one Create card holding status, prompt, Style, Configuration and Generate; Gallery, History, Logs and Quick setup sit underneath, and settings are behind tabs. It can use a manual prompt, a chat scene or supported image tags. Prompt preparation may also call a text model. When a separate connection profile writes those prompts, its Max Tokens defaults to 32,000 and goes up to 64,000, so long prompts aren't cut short; an install still on the old 500 default moves to the new one once. Character Expressions uses separate sprite artwork and classification settings; a character avatar alone isn't a complete expression set.
With the GPT Image (OpenAI) provider, Model accepts a model ID the proxy provides, and Proxy URL points at that proxy. When your proxy offers gpt-image-2.6-sunburst or gpt-image-2.6-flare, those names are sent through its Responses image route, as it already did for the 2.5 names, so they generate a picture instead of failing on the plain images endpoint.
Speech, captioning, translation, attachments and galleries each have their own settings. A working text connection doesn't automatically configure all of them.
Images in replies¶
Settings → Appearance → Avatar & Chat Styles holds Don't send chat images to the model, which keeps image files and their saved descriptions out of new automatic replies across the account. Pictures stay visible and existing message text stays text. Review chat images opens the review panel for the open chat, where an image can be Send the image, Show only or Use a description and, in Roleplay, Remove from message. The account setting and a chat's or branch's own setting override the individual choices.
Quick Image Gen's Generated images in replies chooses what a new picture does: Show only, Use a description or Send the image. Use a description sends the prompt that made the picture, as text. A new Roleplay picture defaults to Show only; a Conversation picture defaults to Send the image. Library images in replies defaults to Send the image and offers Use a description, which sends the library image's saved description. Sending an image also needs Send inline media and a connection that accepts images.
See Roleplay for the per-message controls, Conversation for its review panel, and Something went wrong for repair and recovery.
Separate expression sets for a shared card¶
Open a chat, then Characters & Lore → Character Expressions. The image list is the Expression images section. Under Character in this card, use Add character to give each person in a multi-character card a separate expression set. Whole card selects the shared set. This is separate from group-chat membership and the card's avatar.
Under Character appearance, enter Appearance notes for image generation, then Save appearance. Follow the first named speaker in each reply picks a set from explicit speaker names or Name: dialogue labels, not an ordinary mention. If no name matches, it uses your chosen set. Selecting a set manually turns following off. Remove character set removes the choice from the card but leaves its image folder.
Use Add expressions for several labels separated by commas or new lines. Find an expression searches the list, and All, Ready and Missing filter it. A label doesn't create artwork: use Generate missing with Quick Image Gen, or add files under More image tools with Upload character sheet or Upload ZIP. Export ZIP downloads every image in the selected set, even those hidden by the current search or filter, with the original names and formats; it doesn't include the card, description or chats. Image files are named after the expression, with an optional suffix such as joy.png and joy-2.png.
Create expressions for offers Chat character and a Persona: entry for each saved persona. Persona generation uses the saved persona name and description, and every persona keeps its own set, even when two personas share a display name. Choosing an image for a persona sets it for that persona's latest user message and as the starting choice for future messages; older messages keep their saved expression. Clear persona expression clears the choice without deleting artwork. With Classifier API → In-Chat Agent and the Expressions Agent running automatically, a new persona message without a saved expression is also classified in the background against that persona's own set; the character's reply doesn't wait for the result. That classification is a separate request on the Agent's connection. A valid result replaces the hand-picked image on that new message; an invalid result keeps it.
Under How expressions are picked → Image generation and cleanup, Sprite generation mode offers Separate images or Character sheet (auto-split). Separate images devote more pixels to each expression, and generation size still follows Quick Image Gen. Sprite framing offers Bust portrait (consistent close-up) and Full body (consistent standing pose).
Generation background offers Green (recommended), Magenta (for green characters) and White or transparent. Green is the default; choose a colour that doesn't appear on the character. Green and magenta are removed automatically, while Also remove white or other plain backgrounds is a separate switch. Changing this setting doesn't touch images you already have; it applies to the next generation or cleanup pass. Cleanup smooths white fringes, keeps each tile's full edges without clipping the artwork, and can't reliably tell every background apart from the character, so check the result.
AI background removal (optional) downloads a verified anime cut-out model of about 176 MB once and then processes images on the Neconyan server. Enlarge after cleanup (optional) offers Off, 2x larger and 4x larger, using a model of about 2.5 MB in small overlapping sections and keeping transparency. Both start off. Only the model downloads contact the model host; Quick Image Gen still uses your configured provider for new pictures.
More image tools → Redo cleanup reprocesses the existing set and, after a confirmation, overwrites the files, so Export ZIP first. Repeated cleanup with enlargement still on enlarges again. Stop cancels the run, and images that fail are reported as failed rather than cleaned. As measurements from the project's ARM test server, not promises: background removal took around 22 seconds and about 1 GB of memory for a 256 by 205 pixel image, and 2x enlargement of that size took around 21 seconds with a measured peak of 188 MiB including decoding. Generating images can incur provider charges.
Display and choices → Show expressions in phone chat shows each reply's saved expression above its text on phones, in any chat style, including styles that hide ordinary avatars. Ripple keeps its side portrait. The switch is display-only and doesn't reclassify messages.
Find characters or diagnose a problem¶
BotSearcher searches supported card sources. Available filters, accounts and downloads depend on the source site. External-site errors can occur independently of your model connection. A failed JannyAI link import names its reason, such as a missing login on the server or a hidden card definition. Logging in to JannyAI in your own browser doesn't count; use Refresh status or Open JannyAI login window in BotSearcher.
JannyAI card blocks proxies means the creator has hidden the definition and disabled proxy access. Logging in again won't remove that setting. If JannyAI offers a card download, download it there and import the file; not every hidden card offers one.
The JannyAI panel stays inside BotSearcher's workspace on phones and desktop, including its login and status controls.
Debugger provides a diagnostic report and layout snapshot. Reproduce the problem first, then open its report, read it and copy or download it. Prompt Inspector helps inspect supported outgoing requests; those prompts may contain private chat material.
Taro
'Included' means the code is here. You'll still need to configure any external service it uses and check that service's charges.
Original authors, source links and licence details are recorded in the included-tools attribution table and beside the bundled source. Want to make your own tool? Start with Making extensions.