AI Coach
Overview
AI Coach provides conversation, accountability, and actions for your session. It is intended for non-explicit guidance and session management. What it can actually change depends on the permissions you enable, your lock's configuration, and any required device consent.
Get started
- Open AI Coach settings and choose an available model/provider under Model Setup.
- Choose a Personality and Intensity Level. Intensity sets action limits; it is separate from the coach's tone.
- Add your AI Instructions, then review Enabled Features. Enable only the actions you want the coach to use.
- Enable AI Coach, save your configuration, and open its conversation. If initialization or a message fails, check the provider settings and the error shown in the app.
- Try the configuration before enabling Hardcore Mode, which locks protected settings.
Use Personality for tone, attitude, and speaking style. Use AI Instructions for preferences such as “assign no more than three tasks a day” or “keep tasks around ten minutes.” These instructions guide the model; use the app's actual permission switches and limits for restrictions that must be enforced.
Chastify supplies the available actions automatically. You do not need to copy its system prompt or command examples into your own instructions.
Customize behavior rules
In Persona & Instructions, type directly in Behavior rules to replace the built-in preferences about replies, rewards, refusals, and tone. Bot Coaches have the same editor beside their personality and AI instructions. Owners can edit both public and private bots. Preset bots require an editable copy.
The editor lists the behavior preferences, including response style, rewards, refusals, visibility, hygiene, Hardcore responses, memory, parental control, nudges, and farewells. Edit or replace them, for example: “Be cooperative and direct. Ask questions when useful. Never mock me or treat disagreement as manipulation.” Save the coach to apply your changes to subsequent replies. Clear behavior rules leaves no additional behavior preferences; the configured personality still applies. Restore defaults returns to the built-in behavior. Expand Always-enforced action requirements to review the fixed action restrictions.
Behavior rules cannot grant permissions, bypass device consent or confirmation, change action limits, or disable protected Hardcore Mode. Command validation remains enforced by the server. Behavior editing is locked under the existing Hardcore Mode restrictions.
Tone, teasing, and mocking remain editable. Fixed consent rules take priority: the coach must treat a clear stop request, safeword, consent withdrawal, or report of distress as a real boundary, never a reason to punish or pressure the wearer. An active app session does not prove that someone is physically wearing a cage; conflicting reports should be acknowledged without assuming manipulation.
Device stop actions remain available even when permission or consent to start effects is absent. When a response includes a stop, the server executes only the stop actions and discards other actions in that response. It attempts each applicable device family and reports incomplete stops; an accepted request does not confirm that the physical device stopped. Use the app/device's direct stop controls if stopping cannot be confirmed.
A device stop affects current and queued effects. It does not by itself change saved consent, remove future schedules, disable recurring automation, or end a lock. Use the relevant consent, schedule, and automation controls for persistent changes. Recognition of a stop or withdrawal written in chat still depends on the model; command validation and stop exclusivity are enforced by the server.
See Coach Memory & Proactive Nudges to review remembered details, back up or reset memory, and configure messages while you are away.
To listen to replies, set up Coach voice. You can generate audio for individual messages or enable automatic reading of new replies.
Capabilities
| Capability | What to check before enabling it |
|---|---|
| Chat and memory | Choose a persona and instructions. Memory and chat management are available in settings, subject to Hardcore Mode. |
| Voice and read-aloud | Choose a speech model, voice, and funding source. Manual playback and automatic replies share settings with Bot Coach. See Coach voice. |
| Time and cards | Enable Time Adjustment for adding/removing time or supported card changes. The active lock and intensity limits still apply. |
| Tasks, writing tasks, and quizzes | Configure saved tasks and enable generated tasks separately. Cancellation is its own permission. |
| Hygienic unlocks | Granting a temporary override and changing the regular hygienic-unlock settings are separate permissions. See Hygienic Unlock. |
| Photo verification | Vision capability and verification settings determine the review path. See Task Photo Verification. |
| Calendar access | Off by default. Lets the coach read visible calendar entries and create, edit, or delete entries you can manage. Protected keyholder events remain protected. See Calendar Events & Tasks. |
| Shock and vibration | Starting effects requires the device-command permission, explicit consent, and a supported available device; stopping does not require start permission. See Shock Delivery & Queues for pending and scheduled actions. |
| Android Device Control | Requires the Parental Control feature, phone-control coach access, and an active self-control session. See Self-Control & Release Locks. |
| Pillory and lock settings | Only supported changes can run, within the relevant permissions and lock state. Existing lock pillory settings take priority over AI defaults. |
| Programmable Lock | The active program must allow the function or variable, and the coach needs the permissions required by its actions. |
| Proactive Nudges | Can initiate messages after inactivity. Review the timing, quiet hours, and separate permission for commands during nudges. |
Supported Actions
Ask for actions in ordinary language. A response may contain conversation, an attempted action, or both. Chastify validates the requested actions before applying them. Disabled features, missing consent, unsupported devices, or the current session state can prevent a request from succeeding.
Check the action result, current lock state, or lock history when confirming a change. A sentence such as “I added time” is not by itself evidence that an action succeeded.
For proactive messages, see Commands during nudges for the additional action limits, settings changes, and what happens if you do not reply.
Sensitive changes can require a separate confirmation in the app. For example, an AI request to disable emergency release or future keyholder control is not the same as your confirmation. Review the consequence before confirming.
For Android self-control, Temporary Restriction Bypass lets the coach request a wearer-approved 60-second bypass. This permission must be enabled before Device Control starts; enabling ordinary Parental Control alone does not grant it.
Phone Control also has an AI Coach & Bot Coach parental-control access switch. It must allow access for coach restriction commands to work. Starting a self-control release lock requires confirmation, but an authorized coach can extend an existing fixed release timer without a new confirmation. Review coach access before locking self-control.
Advanced: JSON Action Format
The model returns structured action data alongside its reply. Chastify generates the instructions for that format and checks the actions before execution. You do not need to write JSON to use AI Coach, and pasting internal action examples into your instructions does not grant extra permissions.
If the coach describes a change that never appears in the app, check its enabled features and any reported action error. Avoid adding legacy brace or colon commands to the prompt as a workaround.
Bot geofence context
With geofencing enabled on an active lock, Bot Coach automatically receives a short summary of the latest reported zone entries and exits, such as entered "Home" 2m ago. No extra setting or Programmable Lock is needed. The summary contains at most three crossings from the latest report, with zone names shortened to 48 characters, and expires after 24 hours. Reports involving a location privacy zone are excluded.
The bot sees this context on its next chat response or regular proactive message; a crossing does not start an extra AI request. These are past reported crossings, not proof of the wearer's current location. Coordinates and maps are not included. Disabled or suspended geofencing and ended sessions provide no geofence context.
Voice & Read-aloud
Coach voice turns coach messages into spoken audio. AI Coach and Bot Coach share the same voice settings and saved-audio allowance. Availability depends on whether the feature is enabled for your account; the panel explains if a subscription is required.
Set up a voice
- Open AI Coach settings and expand Coach voice. You can also open Coach voice settings beside a message's audio controls.
- Turn on Enable coach voice. This switch saves immediately.
- Choose how to fund generation. Use your own OpenRouter key, or select Subscriber voice allowance if it is offered. For a personal key, choose the saved Account, Bot Coach, or AI Coach key source. You can enter an account key in this panel.
- Choose a Speech model and Voice from the available options. Under Speaking style, select a preset or enter Style instructions of up to 500 characters, such as “speak warmly and calmly.” These instructions change the audio delivery; the coach's personality and action permissions remain separate.
- Choose Save voice settings before generating audio. Listen to voices for free on Google opens prerecorded samples. Generate custom preview (paid) uses your saved voice and style settings and consumes the selected funding source.
Speech uses a separate model from your chat model. Message text and any style instructions are sent through OpenRouter for speech generation, including when you use a self-hosted chat model.
Play messages and automatic replies
On a coach message, choose Generate audio and wait for generation to finish, then choose Play audio. The player offers pause and Download audio to save an MP3. Playing an existing recording does not generate it again.
For hands-free replies, enable Automatically generate and play replies in Coach voice settings. It is off by default and saves immediately. New replies to messages you type to either coach generate paid audio automatically. Proactive nudges and other automated messages are excluded.
Keep the conversation open and visible for automatic playback. Browsers can require a tap on Play audio, so enabling this option does not guarantee that every reply will start playing unattended. You can play saved audio manually afterward.
Turning off automatic replies prevents new automatic generations; work already generating may finish. Turning off Enable coach voice also disables automatic replies and hides the message audio controls, while keeping saved recordings.
Costs and voice allowance
New recordings, replacements, and custom previews consume OpenRouter credits or the selected subscriber voice allowance. Replaying saved audio is free. Check the current model pricing and the allowance panel's usage and reset date; voice funding is configured separately from chat access.
When Subscriber voice allowance is selected, generation can automatically use the selected personal OpenRouter key after the allowance is exhausted or the allowance entitlement becomes unavailable. Without that key, generation stops. Pending usage accounting or a temporary server-generation outage can instead require you to wait or explicitly select your own key.
A failed request may already have incurred a charge. Read the confirmation before choosing Retry generation; another attempt can charge again. Deleting a recording does not refund its generation usage.
Saved audio and replacements
Expand Saved audio in Coach voice settings to play, download, or delete recordings and previews. Delete audio removes the recording while keeping the text message. Clear saved audio removes the library.
Changing the voice or speaking style affects future generations. Existing recordings keep their original voice. Replace audio creates a new recording using the current settings and funding source; the existing recording stays until its replacement is saved. An edited message can show a notice that its recording uses an earlier version.
The current storage limit is 100 MiB or 200 clips, whichever is reached first. Each generation needs 10 MiB of free space while it is being prepared. Delete unwanted recordings if the library is full. Saved audio is not a permanent archive: deleting its message, automatic chat trimming, or account deletion also removes the associated recording. Download anything you want to keep.
If audio does not work
- No audio controls: check feature availability, enable Coach voice, and save a supported model and voice with an available funding source.
- No automatic sound: keep the conversation visible, check device volume, and tap Play audio if the browser blocks autoplay. Nudges are not automatically read aloud.
- Generation is busy or storage is full: wait for the current generation to finish, leave a few seconds between requests, or free space in Saved audio.
- A message is too long: speech accepts up to 4,000 characters after formatting is removed. Ask the coach for a shorter reply.
- A key, allowance, or model error appears: follow the panel's message, update the selected key or voice settings, and save before retrying. Use Reload audio status when it is offered.
Model Setup
The setup page shows which provider choices are available to your account. Available models, membership benefits, promotions, and usage limits can change, so use the current model picker and usage summary rather than an old list of model names.
OpenRouter
If OpenRouter is available, enter your own API key and select a model from the picker, or enter its model ID. Chastify saves the key encrypted and uses it for provider requests. Usage is charged against your OpenRouter account rather than Chastify's included AI allowance.
Choose a model with image-input support if you want to send images or use AI photo review. Check the capability information in the picker. When AI Plan routing is on, turn it off before configuring your own provider.
AI Plan
Eligible memberships can use hosted AI through AI Plan without entering their own provider key. The current plan panel shows available models, usage, and any applicable limits or estimates. See Support the Site for membership options.
A model advertised with a temporary promotion may later change availability or allowance. Check the panel before switching models or relying on unlimited usage.
Demo AI Mode
When enabled for your account, Demo AI provides a server-selected model without your own API key. It is available in AI Coach as well as supported bot-chat flows. It is not a permanent entitlement.
The setup page shows whether Demo AI is available and any daily message allowance, messages used, or rate-limit message. A complete personal OpenRouter configuration can take precedence over Demo AI. If the demo is unavailable or its allowance is exhausted, choose another provider available to your account.
Self-Hosted (LM Studio / Ollama)
When Local Model is available for your account, you can use a server that exposes a supported chat-completions API, including an appropriately configured self-hosted model.
- Run your model server and confirm its model ID and chat-completions endpoint.
- Make that endpoint reachable from Chastify's server through a public URL. A local
localhostor private-LAN address will not work in the hosted app. - Enter the full endpoint in Local Model URL, for example
https://your-model.example/v1/chat/completions, and enter the model ID. - If the endpoint requires authentication, enable Use API Key for Local Model and enter the key. Protect a publicly exposed endpoint with authentication.
- Enable Visual support only when both the endpoint and model accept image inputs. Save and test a message.
The model server must remain reachable while you use it. If you use a temporary tunnel, update the saved URL when the tunnel address changes. Self-hosting does not turn this into offline chat: Chastify's server still processes the conversation and calls your endpoint.
Dice, wheel, and task module settings
AI Coach and Bot Coach can enable, disable, and edit these modules during an active lock when their corresponding permission is enabled:
- Dice Module Control: dice types/counts, roll frequency, time multiplier, unfairness, and freeze when ready.
- Wheel Module Control: wheel options and outcomes, spin cooldown, behavior while frozen, and editing permissions.
- Task Module Control: saved tasks, cooldown, selection mode, point requirements, automatic assignment schedules, failure outcomes, and editing permissions.
All three permissions are off by default. Enable them in AI Coach settings or in your Bot Coach's editor. Task Assignment and Custom Task Assignment alone do not grant permission to change the saved task module. Task-editing permissions do not depend on Hygienic Unlock Settings.
Enabling a module or adding/changing its outcomes also requires the relevant coach permission: Time Adjustment, Device Control, Pillory Control, Task Assignment, or Photo Verification. Existing device consent and lock limits still apply. A new wheel needs at least three valid options. Linked writing tasks must already exist and belong to the wearer, lock creator, or keyholder; the coach cannot remove an existing task's exclusion from AI assignment. Untrusted shared-keyholder restrictions on task consequences still apply.
Changes to cooldowns, wheel option counts, or scheduled task pools also require the permissions for the outcomes they affect. If someone edits the same settings while a coach is preparing a reply, the conflicting coach change is rejected; ask the coach to read the latest settings and try again. Changes to unrelated settings are preserved.
For example, with Dice Module Control and Time Adjustment enabled, ask: “Enable dice with two six-sided dice, a two-hour cooldown, and a one-minute multiplier.” With Task Module Control enabled, ask: “Add a saved task called Evening walk, leaving the other tasks unchanged.”
Disabling a module preserves its configuration. It does not cancel existing task runs or undo an active freeze. Scheduled task assignments pause when the task module is disabled. The existing request to stop dice freeze-on-ready remains available separately. Module changes are unavailable in between-locks chat; a proposed future session must still be reviewed and started by the wearer.
Session names, pillory names, and penalties
AI Coach and Bot Coach can now change three settings during an active session:
- Session Name Control lets the coach rename the session, including the name shown on public voting.
- Pillory Control lets the coach rename an active pillory's reason without restarting its timer, clearing votes, or changing its link.
- Penalty System Control lets the coach enable or disable the already-configured Penalty System. It preserves quotas, periods, and punishment settings. Enabling requires Time Adjustment and, if the saved penalties include pillory, Pillory Control. Enabling penalties on a shared lock also requires a trusted keyholder.
Session Name Control and Penalty System Control are off by default; enable them in AI Coach settings or the Bot Coach editor. Existing Hardcore Mode permission restrictions apply. These controls are unavailable between locks.
For example: “Rename this session to November challenge”, “Change the pillory reason to Missed check-in without extending it”, or “Disable the Penalty System while I am away.”
Disabling the Penalty System does not undo penalties already applied or disable separate task, puzzle, game, or verification punishments. Re-enabling uses the existing Penalty System scheduling rules. Calendar access does not automatically suspend or restore penalties: ask the coach to enable them again when you return. This does not add one-puzzle rewards, public voting configuration, control-pass copying, discoverability changes, or human/bot keyholder handover.
Puzzle and game availability
Enable Puzzle and Game Control in AI Coach settings or the Bot Coach editor. It is off by default and unavailable between locks. This permission also governs the existing individual casino toggle commands.
The coach can enable or disable saved Jigsaw, Daily Puzzles (Wordle, Sudoku, and Hangman together), Flappy Bird, and Simon Says extensions, plus chess, the casino master switch, and individual Blackjack, Slots, Roulette, Plinko, Crash, and Poker games. Configure puzzle/arcade extensions on the lock first; the coach cannot install them, upload images, or change their configuration through this control. Dice, wheel, and task modules keep their separate permissions.
For example: “Disable Daily Puzzles and Simon Says while I am away” or “Enable my configured Jigsaw puzzle.” Enabling also checks the permissions for saved time, device, pillory, task, or hygienic-unlock outcomes, extension availability, and subscription access. Shared locks require a trusted keyholder to enable games. Enable the casino master switch before enabling an individual casino game when the whole casino is off.
Toggles preserve configuration, progress, history, counters, and cooldowns. Disabling prevents new play and disabled-extension requirements from running; it does not cancel started games, reset their deadlines, undo applied penalties, or generally remove freezes. Daily Puzzles reconciles its own availability freeze through its existing lifecycle. Re-enabling follows existing game scheduling rules. Automatic one-puzzle rewards and automatic holiday suspension/resumption are not included; ask the coach to toggle the game again when needed.
AI Tasks
| Task type | Setup |
|---|---|
| Saved task | Enable Task Assignment, then use Configure available tasks for the active lock. |
| Generated task | Enable Custom Task Assignment. These one-off tasks are not saved to your task module. |
| Writing task or quiz | Enable Ad-hoc Writing Tasks & Quizzes. Set the maximum typing repetitions before enabling Hardcore Mode. |
| Cancel an active task | Enable Cancel Tasks separately. It can also cancel a task assigned by a keyholder, wheel, or extension. |
Require photo verification for ad-hoc tasks affects generated tasks; saved tasks use their own photo-verification setting. The destination depends on vision support and the lock's verification policy. Check the review destination shown in the task panel, and read Task Photo Verification before enabling photo requirements.
Strict verification
Enable Strict verification near Enabled Features to require the AI Coach to check a verification code in uploaded task proof and cage-check photos. Bot Coach offers the same option beside the safety profile when you activate a bot for a lock session. Strict verification is required while that coach's Hardcore Mode is active.
Open the task or cage-check panel to see its code. Write it on paper and keep it clearly readable in the photo alongside the requested proof; typed captions and added overlays do not count. For before/after tasks, include the code in each uploaded photo. While camera attestation enforcement is disabled, in-app camera photos can omit the code, including photos from older native apps. When enforcement is enabled, only camera captures verified by the server are exempt. The verification panel shows the current camera requirements. The coach still checks whether each photo satisfies the saved task or cage-check instructions. Task generation and chat follow the same code policy. For timed tasks, the reviewer uses the server timer state when the photo was submitted; it must not demand timestamps or extra before/after photos unless the saved task requires them.
For example, when uploading a cage-check photo, include the displayed six-digit code on paper in the same scene as the cage. Submit proof from its verification panel rather than chat. Existing pending cage checks show their code when you reopen the panel after enabling strict verification.
Hardcore Mode
AI Coach's Hardcore Mode keeps AI enabled during an active lock session. It locks persona/instruction editing, intensity, enabled permissions, and memory/chat management. Already-enabled features must stay on. Some features that are off cannot be added afterward, including Time Adjustment and Hygienic Unlock Grant.
You can turn it off when there is no active lock session. This AI Coach setting does not offer a 24-hour or seven-day expiry. Bot Hardcore Mode is a separate session feature with its own duration and restrictions; read its activation confirmation.
Before activating either coach's Hardcore Mode, choose Allow deleting recent messages. It is unchecked by default to protect recent chat history. Check it if you want to remove the latest message turn to recover from unwanted or repetitive replies. The choice is locked in both directions while Hardcore Mode is active, including when extending Bot Hardcore Mode. Existing Hardcore sessions keep recent-message deletion available. Full-history clearing and memory changes remain restricted regardless of this choice. Deleting a message does not undo actions the coach already performed.
Command Limits
AI Coach uses the Soft, Medium, and Hard intensity levels shown in its settings. The time-addition budget covers the preceding 24 hours, rather than resetting at midnight.
| Action | Soft | Medium | Hard |
|---|---|---|---|
| Add time | 12 hours total per 24 hours; up to 1 day per action, subject to that budget | 24 hours total per 24 hours; up to 1 day per action | No daily addition budget; up to 31 days per action |
| Remove time | Up to 12 hours per action | Up to 12 hours per action | Up to 31 days per action |
| Add/remove cards | Up to 3 supported cards per action | Up to 3 supported cards per action | Up to 3 supported cards per action |
| Freeze | Up to 1 day per action | Up to 1 day per action | Up to 1 day per action |
The remaining time, configured minimum and maximum durations, enabled modules, and other session rules still apply. The coach cannot add or remove green cards through ordinary card-management actions. Verification and task requests do not have a general fixed daily action cap; their own feature and task restrictions still apply.
Separate Bot Hardcore Mode changes some action limits, including the daily addition budget and freeze-duration restriction. Choosing Hard intensity or AI Coach's own Hardcore switch is not the same as activating that bot session mode.