Help Centre

The fictional sample, drafting and local calculations need no API keys and make no model calls. Explore a review in five steps, without provider API keys.

Pressure-test this starts with a short plan or answer. Compare options uses explicit alternatives and criteria. Build a reviewed brief turns source material into a reusable document.

Open a finding to inspect its exact saved source passage. Disputed, unverified and user-supplied findings remain distinct. A quotation shows provenance, not proof. Model agreement, matching source excerpts and unresolved questions are shown separately. Agreement and a matching quotation do not prove factual correctness.

Add your own ratings from 0 to 10 and set criterion weights. Weighted comparisons update locally without a model call. Missing ratings stay missing; the results reflect your judgement. Changing sources, options or priorities marks earlier AI analysis stale. A confirmed new review creates a proposal to compare in What changed. Your edited brief remains separate.

A full review previews two separate assessments and one synthesis call. A targeted challenge previews one call. Confirm the sources, models, route and limits before each run. Review the named route before confirming. An approved review uses that route and does not silently switch providers after a failure.

Import supported documents, drop files or send text from other apps. On Mac, Ask AI Council and Check this answer with AI Council in the Services menu start a council conversation with the text, and Pressure-test with AI Council saves a review draft. On iPhone and iPad, Prepare a review in the Share sheet offers the same choices. App Shortcuts offer these actions too. Nothing is sent to a model until you send the message or start the review. Shared URLs are saved without fetching the page. Extraction limits and image omissions stay visible. Edit your recommendation and export Markdown, JSON or a printable PDF with source and revision context. Inspect the readable export before sharing it.

Saved work remains available after an interrupted review. Inspect the unfinished run and its preview before choosing to resume; reopening a workspace does not make a new model call.

Council, Direct Chat, Comparison, templates, themes and saved conversation history remain available. Use Review this conversation to make a separate snapshot. The original stays unchanged; saved model answers are not verified source facts.

Stage 1: Gathering Responses
Your question is sent to multiple AI systems at the same time. Each AI provides its own independent answer without seeing what the others have written.

Stage 2: Peer Review
Anonymous labels hide model names during peer review. They do not eliminate bias or guarantee correct answers.

Stage 3: Synthesis
A designated "Chairman" AI reads all the original responses and the peer reviews, then produces a final answer that incorporates the best insights from everyone. This synthesis is highlighted so you can easily find the conclusion.

Open Settings (Command+,) and go to the Providers tab. Choose Configure beside a provider to add its API key. The provider button loads available models; OpenRouter also checks the saved key. No inference request is sent.

Popular providers to get started with:

• OpenAI: platform.openai.com/api-keys
• Anthropic: console.anthropic.com
• Google: aistudio.google.com/apikey
• xAI: console.x.ai
• OpenRouter: openrouter.ai/keys

API keys are stored securely in your Mac's Keychain.

In 8.0Mac App StoreiPhone and iPadSetappAvailable now on Setapp. The App Store versions are awaiting Apple's review.

Use a custom provider for a self-hosted server or a third-party service that follows the OpenAI chat format, such as a vLLM server or a gateway you run yourself.

1. Open Settings, then Providers, and choose Add Custom Provider.
2. Enter a name. It appears in settings and in model pickers.
3. Enter the base URL, including the version path if the service needs one, for example https://api.example.com/v1. Use https://, an IP address or a .local name, because Apple platforms block http:// with a host name.
4. Add an API key if the service needs one, or leave the field empty for a server without keys. The key is saved only when you choose Save.
5. Choose Test connection. It calls the address once with the key you typed and reports the HTTP status or how many models it found. Nothing is saved yet.
6. If the server lists no models, or does not offer a model list, type the model IDs in the Model IDs field. Separate IDs with commas or new lines.
7. Choose Save, then open Settings, Council and add the provider's models to your council.

To change a provider later, choose Edit beside it in Providers. Its route is kept, so saved councils that use it keep working. On iPhone and iPad, localhost means the device itself, so use the server's IP address or a .local name.

Local-only review blocks cloud model requests and web search. Supported local models use device resources; availability depends on the device and model setup.

For supported Mac local models, go to the Local Models tab in Settings:

Download from Featured or Hugging Face:

• Browse the Featured section for recommended models, or use the Search Hugging Face tab
• Check the compatibility badge to ensure the model fits your Mac's RAM
• Click Download. New models are saved in Application Support/AI Council/Models
• To save new downloads somewhere else, choose Change under Downloads Folder. Models you already have stay where they are and are not moved.
• Before a download starts, AI Council checks the free space. If there is not enough, it tells you how much the download needs and how much is available
• Cancel stops a download that is running. If a download is interrupted, Resume carries on where it stopped when that is possible. Otherwise choose Retry
• The model will appear in model selection dropdowns throughout the app

Link Existing Models:

• Go to the My Models tab
• Click "Link Model" to browse for GGUF files already on your Mac
• Select any .gguf file to add it to AI Council without copying

Remove or hide models:

• Delete removes a downloaded model from disk and frees its space. You confirm first, and it cannot be undone
• Unlink removes a linked model from AI Council only. The file stays where it is and is not deleted
• Hide keeps a model out of the model selectors. Models you hid are listed with a Restore button each, and Restore All brings them all back

Use Ollama Models:

• Ensure Ollama is running on your Mac
• Go to Settings > Council tab
• Click "Add Model" to open the model picker
• Ollama models appear at the top under "Ollama (Local)"
• Click any model to add it to your council

In Settings, go to the Council tab to choose which AI models participate in your council:

• Click "Add Model" to open the model picker
• Browse models grouped by provider, or use search to filter
• Click any model to add it to your council
• Remove models by clicking the red minus button next to them
• Choose which model serves as Chairman for final synthesis
• Optionally save your configuration as a preset

The status indicators show model availability:

• Green: Direct API connection available
• Orange: Available via OpenRouter fallback
• Blue: Local model (Ollama or downloaded GGUF)
• Red: No API key configured for this provider

In 8.0Mac App StoreiPhone and iPadSetappAvailable now on Setapp. The App Store versions are awaiting Apple's review.

1. In Settings, Council, choose the members and the chair you want.
2. Choose Save Current, give the council a name and, if you like, a description.
3. If the name matches a saved council, the sheet says it replaces that council. The saved council is updated in place. To keep both councils, choose Save as New. The copy is saved with (2) after its name. Names of the built-in presets cannot be reused.
4. Apply a saved council by choosing it in the preset list.
5. Rename, Edit and Delete sit beside each saved council. Edit changes the saved lineup only and does not change the council you are using now. Update from Current Council copies your live council into the saved one.

Ready-made presets show only models from providers you have set up. Saved councils stay in the list even when a provider is not available.

A saved council holds up to 5 models. To add a model that is not in the list, choose Add by model ID at the foot of the Add Model list.

If a provider stops listing a model you selected, AI Council selects the closest model from the same provider and shows a Model Updated banner. If there is no close match, the model stays selected and a Model Not Listed banner explains it. A model from another vendor is never substituted. You can change either choice in Settings.

Choose a Setapp route to use managed AI on Mac without provider API keys. Availability depends on your Setapp account, credits and selected model. You can also configure your own provider accounts.

The fictional sample, drafting and local calculations need no API keys and make no model calls.

Review the named route before confirming. An approved review uses that route and does not silently switch providers after a failure.

• Click the plus button or press Command+N to start a new conversation
• The new conversation appears in the sidebar with a smooth entrance animation
• A welcome view appears with helpful tips for getting the most from the council
• The input area pulses gently to draw your attention
• Type your question. On Mac, press Return to send and Shift+Return to add a line. On iPhone and iPad, use the Send button.
• Press Command+period to stop an answer while it is running.
• If you stop before an answer is saved, your question stays in the conversation with Try Again, Edit and Discard.
• Paste or drop an image into the message box, then add a question about it. An image is sent with a question.
• Watch as the three stages complete and your synthesised answer appears

AI Council supports cloud AI providers, including:

• OpenAI: GPT and o-series reasoning models
• Anthropic: Claude Sonnet, Opus and Haiku
• Google: Gemini Pro and Gemini Flash
• xAI: Grok models
• Mistral AI: Mistral Large, Mistral Medium
• Groq: Ultra-fast inference for open models
• Together AI: Open-source model hosting
• Perplexity: Search-augmented AI
• DeepSeek: Specialised coding models
• Cohere: Enterprise-focused models
• Cerebras, SambaNova, Fireworks, DeepInfra, MiniMax, Z.ai
• OpenRouter: Access to models available through OpenRouter
• Custom: Any OpenAI-compatible endpoint, such as a self-hosted server

Model lists refresh from the providers you have set up, so new models appear without an app update. Add any other OpenAI-compatible service as a custom provider. Configure providers in the Providers tab of Settings.

AI Council refreshes model catalogues automatically. Review suggestions use cached capability information and need explicit Apply. Availability can change; saved review choices are not proof of current access.

Local-only review blocks cloud model requests and web search. Supported local models use device resources; availability depends on the device and model setup.

• Metal GPU Acceleration: Optimised for Apple Silicon (M1/M2/M3/M4) performance
• Local-only review: Cloud model requests and web search are blocked
• Offline Operation: Works without internet connection
• Zero API Costs: No per-query charges after download

The Local Models tab in Settings has three sections:

• Featured: Curated list of popular models from trusted sources (bartowski, TheBloke, and others) with compatibility indicators showing if they fit your Mac's RAM
• Search Hugging Face: Search Hugging Face for GGUF models, filter by quantisation level, and download with progress tracking
• My Models: View and manage your models, link existing GGUF files from anywhere on disk, and scan for models from Ollama, LM Studio, or GPT4All

Each model shows a compatibility badge based on your Mac's available RAM, so you know before downloading whether it will run well.

Note: Ollama integration is available on macOS only and is not available on iOS.

If you have Ollama installed on your Mac, AI Council can use your Ollama models directly:

• Auto-Detection: AI Council automatically detects when Ollama is running
• Model Picker: Click "Add Model" in Council settings - Ollama models appear at the top under "Ollama (Local)"
• No Configuration: Ollama models work immediately without API keys
• Blue Indicator: Local models show a blue status dot in the council configuration

This is ideal if you already use Ollama and want to include those models in your council without downloading them again.

Get started quickly with pre-built prompts for common tasks:

Coding

• Code Review: Have your code examined for bugs, security issues, and improvements
• Debug Help: Get assistance with error messages and problematic code
• Architecture Design: Plan the structure of your software projects

Writing

• Proofread: Check your text for grammar, clarity, and style
• Summarise: Get a concise version of longer documents

Research

• Compare Options: Evaluate multiple choices with pros, cons, and recommendations
• Explain Concept: Get clear explanations of complex topics

Analysis

• SWOT Analysis: Examine strengths, weaknesses, opportunities, and threats
• Risk Assessment: Identify and evaluate potential risks

Select a template from the new conversation menu to start with a structured prompt.

Choose how the council members interact with each other. Different modes suit different types of questions:

• Standard: Models answer independently in parallel (default)
• Devil's Advocate: One model actively challenges assumptions and finds flaws
• Expert Panel: Each model adopts a specialist role (Security Expert, Performance Engineer, UX Designer, Software Architect, QA Engineer, Business Analyst)
• Debate: Models debate in rounds before the chair synthesises, responding to each other's arguments
• Consensus Seeker: Models focus on finding common ground and agreement

Select your preferred mode in the Council tab of Settings. The mode indicator appears in the chat interface.

After the council deliberates, AI Council analyses the responses to show you:

• Confidence Score: A heuristic comparison of model responses, not a probability that the answer is correct
• Consensus Points: Key points where all models agree
• Disputed Points: Areas where models disagree, showing different perspectives

Inspect the synthesis and its sources before relying on a claim. The legacy confidence score is an agreement heuristic.

Choose what matters when models evaluate each other's responses. Five built-in presets are available:

• Standard: Accuracy, completeness, clarity
• Coding: Correctness, performance, maintainability, security
• Creative: Originality, engagement, coherence, style
• Research: Rigour, evidence, balance, depth
• Decision Making: Practicality, trade-offs, recommendation, risk assessment

Select criteria in the Council tab of Settings. A preview shows exactly what the peer reviewers will evaluate.

Organise your work into projects with persistent context. Each project can have:

• System Prompt: A custom instruction that's automatically prepended to every query within the project
• Scoped Conversations: Conversations are grouped by project, making it easy to find related discussions
• Project Overview: Name and description to help you identify each workspace

Create projects from the sidebar using the folder button. Switch between projects or view "All Conversations" to see everything. The active project's context is applied automatically to all new queries.

Choose how the chair formats the final synthesis to match your needs:

• Standard: Balanced analysis with natural structure
• Executive Summary: Concise 3-5 bullet points with key takeaways
• Detailed Analysis: Comprehensive sections with thorough explanations
• Comparison Table: Side-by-side evaluation of options with trade-offs
• Action Plan: Numbered steps with clear next actions

Select your preferred style from the controls panel above the message input. The chair's prompt is automatically adjusted to produce the requested format.

Templates suggest sections for a document; review the generated output before using it:

• Product Requirements Document (PRD): Overview, requirements, user stories, success metrics, timeline
• Code Audit: Executive summary, critical issues, security concerns, recommendations
• Meeting Notes: Key decisions, action items, discussion summary, next steps
• Business Case: Problem, proposed solution, costs, benefits, alternatives, risks, recommendation
• Competitive Analysis: Market overview, competitor profiles, feature comparison, pricing, SWOT, recommendations

Select a deliverable template from the controls panel. It adds instructions for the synthesis; check the resulting structure and content before exporting.

Execute multi-step workflows where each stage feeds into the next:

• Research Report: Gather Information > Analyse Findings > Write Report
• Decision Process: Brainstorm Options > Evaluate Options > Recommend > Create Action Plan
• Code Review & Refactor: Identify Issues > Propose Fixes > Refactor Code
• SWOT Analysis: Internal Analysis > External Analysis > Strategic Recommendations
• Content Creation: Create Outline > Write Draft > Refine & Polish
• Problem Solving: Define Problem > Root Cause Analysis > Develop Solutions

Select a chain from the controls panel before sending your query. The chain progress view shows your current step and allows you to continue to the next step or end the workflow.

Evidence Mode asks for citations. Check that each reference exists and supports the claim.

• Toggle Evidence Mode when you have attachments
• The council automatically chunks your documents into numbered sections
• The chair is prompted to cite document section numbers
• The prompt asks the model to state when evidence is missing

This is essential for professional document analysis where you need to verify sources.

Record important decisions with full council context:

• Click "Log Decision" on any synthesis card
• Enter the decision outcome and your reasoning
• The decision is saved with the full council context (models, rankings, synthesis)
• Access all logged decisions from the sidebar
• Export decisions as a formatted document for records

This creates an auditable trail showing how AI-informed decisions were made.

Your decision log is stored on this device only. It does not sync through iCloud, so it does not appear on your other devices. Export the log to keep a backup or to take a copy to another device.

Presets provide optimised model combinations for different tasks, and they automatically adapt based on which providers you have configured:

• Balanced Council: Diverse mix of major providers for well-rounded perspectives
• Speed Focus: Fast models (GPT mini and nano, Claude Haiku, Gemini Flash) for quick iterations
• Code Review: Models chosen for code understanding from Anthropic, OpenAI, DeepSeek and Google
• Creative Writing: Models with strong creative capabilities
• Research & Analysis: Models suited for factual analysis and research

Presets intelligently filter to show only models from providers you have API keys for. You can also save your own custom presets for configurations you use frequently.

Chat one-on-one with any single model without invoking the full council. Select a model from the dropdown, which shows all models from every configured provider grouped by name.

• Full Controls: Response depth, synthesis style, deliverable templates, file attachments, and the Prompt Enhancer are all available in direct chat
• Pin a Favourite: Pin a model for quick access at the top of the selector
• Smart Escalation: Tap "Ask the Council" to escalate the current conversation to a full council deliberation with context preserved
• A Response truncated note appears under the answer with Continue and Retry
• Continue asks the model to carry on from where it stopped. Retry asks the same question again

Direct chat is ideal for quick questions that do not need multi-model deliberation. The council becomes a "go deeper" option rather than the only mode.

Web Search augments your queries with live search results from the internet, giving AI models access to current information beyond their training data.

Toggle the globe icon in the action bar before sending your message. When enabled, relevant web results are fetched and included as context alongside your question.

• Search-Augmented Queries: Relevant web results are fetched and included as context alongside your question
• Works in Council, Direct Chat and Comparison: Available in all three modes
• Provider Support: Requires an API key for Brave Search, Exa, Tavily, Kagi, Perplexity, SerpAPI, You.com or Google Gemini
• Add keys in Settings > Web Search. Choose Auto to try your saved providers in turn, or pick one provider. If the provider you pick has no key, Auto runs instead and a notice says so.
• Each search key has Test & Save, Save without verifying and Remove.

This is useful for research queries, current events, or any question where up-to-date information matters.

Attach images (screenshots, diagrams, photos) to your queries for vision-capable models to analyse.

• Smart Visibility: The image attachment button only appears when your selected model (direct chat) or at least one council member supports vision
• Vision Warnings: In council mode, if some members lack vision support, a warning lists which models will skip the image
• Supported Formats: JPEG, PNG, WebP, GIF, and HEIC
• Supported Models: GPT, Claude, Gemini, Pixtral, Qwen-VL, and other vision-capable models

Attach up to 5 images per message. Thumbnails appear above the text input with a remove button on each.

Seamlessly move from a direct chat to a full council deliberation when you need deeper analysis:

• Context Preserved: The entire conversation history is passed to the council, so you do not need to repeat yourself
• One-Tap Escalation: Tap "Ask the Council" at any point during a direct chat
• Full Deliberation: All three stages run with the benefit of the prior conversation context

This lets you start with a quick direct chat and escalate to the council only when you need multiple perspectives or structured peer review.

Send the same prompt to 2 or 3 models simultaneously and view their responses side by side:

• Side-by-Side Columns: Each model's response streams in its own column for easy comparison
• Model Selection: Choose any combination of models from your configured providers
• Real-Time Streaming: Watch all responses arrive simultaneously
• Stop all ends every model that is still answering and keeps what has arrived. Finished answers are not touched
• Retry under a column runs that model again and saves only that column

Comparison View is ideal for evaluating how different models handle the same prompt, comparing writing styles, or benchmarking model capabilities for a specific task. Unlike the council, it shows raw responses without peer review or synthesis.

Save, organise, and reuse your best prompts with a personal prompt library:

• Save Prompts: Create named prompts with tags for easy organisation
• Placeholders: Use {{variable}} syntax for reusable templates that prompt for values when used
• Tag and Search: Organise prompts by category and search across your entire collection
• Quick Insert: Tap any saved prompt to insert it into the message input
• Enhancer Integration: Send saved prompts through the Prompt Enhancer for AI-powered refinement
• Stored on this device: Your prompt library stays on the device where you saved it

Real-Time Streaming: Watch AI responses arrive in real-time as they are generated. Each council member's response streams live with animated progress indicators.

Response Depth Control: Control how detailed you want the council's responses with the depth selector next to the send button:

• Brief: Concise answers that get straight to the point
• Standard: Balanced level of detail (default)
• Detailed: Comprehensive responses with thorough explanations

File Attachments: Attach documents to your questions. Supported formats include PDF, Microsoft Word, plain text, and code files.

Image Attachments: Attach images (JPEG, PNG, WebP, GIF, HEIC) to your questions for vision-capable models. The image button only appears when at least one active model supports vision. In council mode, a warning lists any non-vision models that will skip the image.

A stepped wizard that transforms rough ideas into optimised, structured prompts. Access it from the wand button in the message input bar or with Command+Shift+E.

• Stepped Clarification: The enhancer asks one focused question at a time, with a progress indicator showing your position (e.g. "Step 2 of 3"). This ensures each question gets a thoughtful answer rather than dumping everything at once.
• Model Selection: Choose which AI model drives the enhancement process. Optionally select a target model so the output is optimised for that model's strengths and prompting style.
• Framework Knowledge: Uses over 10 prompt engineering frameworks including RISEN, RADAR, Chain-of-Thought, ReAct, Few-Shot, Self-Consistency, and more. The enhancer selects the optimal combination for your specific prompt.
• Model-Specific Best Practices: When you select a target model, the enhancer applies tailored guidance for Claude, GPT, or Gemini models.
• Skip or Retry: Skip remaining clarification questions to get your enhanced prompt sooner. If an error occurs, retry with the same model or switch to a different one.
• Use Prompt: The enhanced prompt appears in a copyable card. Tap "Use Prompt" to send it directly to the council input, or copy it to your clipboard.

Second Opinions: After receiving the chair's synthesis, request an alternative perspective from a different council member. Click the Second Opinion button on any synthesis card.

Follow-up Conversations: Continue the discussion with the chair without running a new council. Follow-up questions go directly to whoever provided the most recent synthesis.

A follow-up answered by the chair alone is labelled as such, without peer review. It carries an Ask the Council action that runs the same question through the full council.

Health Check: Before sending your question to the council, the app performs a quick health check on all configured models to verify they are responding. Any models that fail are clearly indicated, and the council continues with available models.

Cost Estimation: Before you send a question, AI Council shows an estimate of API costs.

Token Counter: The input area displays an estimate of how many tokens your message contains.

Smart Follow-up Suggestions: After receiving the chair's synthesis, AI Council suggests relevant follow-up questions based on the discussion, highlighting unexplored angles and areas of disagreement.

Search: Press Command+Shift+F for full-text search across all conversations.

Export: Save conversations in Markdown or JSON format with Command+E.

Model Analytics: Track how well each AI model performs over time - query counts, win rates, and average rankings. Access with Command+Shift+A.

• Theme: System (follows your Mac), Light, or Dark mode
• Accent Colour: Blue (default), Purple, Pink, Orange, Green, Teal, Indigo, Red, Cyan
• Font: System (SF Pro), Inter, Roboto, Open Sans, Lato, Nunito, Poppins, Source Sans 3, Fira Sans, JetBrains Mono, IBM Plex Sans, Work Sans
• Font Size: Small (13pt), Medium (15pt), Large (17pt)
• Density: Compact, Standard or Spacious spacing

• Verify your API key is correct in Settings > Providers
• Use the model-list button in provider settings. OpenRouter also checks the saved key. This does not send an inference request.
• Check if you have sufficient credits/quota with the provider
• Some models may be temporarily unavailable - try again later
• For local models, ensure Ollama is running or the GGUF file is valid

The council will continue with available models even if some fail.

A failed step never deletes the answers that came before it. Each of these keeps your work and shows what you can do next.

• This card appears when the chair fails after the members have answered. Retry or Choose Alternative asks a chair again
• Choose Cancel to close the card. The member answers stay in the conversation, and Generate Chair's Report asks the chair again whenever you are ready

• Your question stays in the conversation with Try Again, Edit and Discard
• Stop ends a follow-up that is still running. The question comes back so you can send it again

This means the key that opens your saved data is missing on this device. Nothing has been deleted. To fix it, open Settings, go to the iCloud Sync section and choose Data Recovery. It offers three actions:

• Retry checks again after you turn on iCloud Keychain and wait for it to sync.
• Export What Can Be Read saves a copy of every item that can still be opened, as plain files in a folder you choose.
• Archive Unreadable Data and Start Fresh asks you to confirm first. It then moves the unreadable data into a recovery folder. Nothing is deleted, and you can restore it later if you find the key.

Your copies in iCloud stay unchanged unless you choose to erase them during the archive step.

Note: Local model support is available on macOS only.

• Check the compatibility badge before downloading - it shows if the model fits in your RAM
• Use quantised models (Q4, Q5) for faster inference on less powerful Macs
• Close other memory-intensive applications
• Apple Silicon Macs (M1/M2/M3/M4) perform significantly better than Intel Macs

For the best experience, use models that show a green "Recommended" badge.

• Ensure Ollama is running (check for the Ollama icon in your menu bar)
• Verify Ollama is accessible at localhost:11434
• Click "Add Model" in the Council settings - Ollama models appear at the top under "Ollama (Local)"
• If models still don't appear, try restarting both Ollama and AI Council

Review workspaces and conversations are encrypted in app storage. Cloud requests send the selected material to the named provider or gateway. Exported files are readable documents.

A cloud request sends your question and included source or conversation context to the selected provider or gateway. Each service applies its own data policies.

Optional iCloud sync uses your private CloudKit database. It requires iCloud access and compatible app settings. Local storage encryption does not make exported files private.

The Privacy Policy lists every service the app can connect to and explains how keys sync through iCloud Keychain.

Open Settings > Usage to inspect recorded usage, provider token counts when returned and estimated costs. These records are not your provider invoice.

Request costs and token counts may be estimated. Provider usage is shown when returned. Direct provider prices do not establish Setapp credit charges. Check provider or Setapp account records for authoritative charges.

Use the filter bar beneath the header cards to narrow records by time range, provider, or conversation. Preset ranges are Last 14 Days (default), Last 30 Days, This Month, and All Time. You can also filter to a single provider to see what a specific API key is costing.

The stacked bar chart below the filter bar groups usage by day. Each segment colour represents a different provider, making it easy to spot which service dominated a given day. Hover over a bar on macOS to see the day's breakdown.

Tap Export CSV to save a comma-separated file of all visible records. The file follows RFC 4180 and includes columns for date, provider, model, prompt tokens, completion tokens, estimated cost, and conversation ID. You can open it in Numbers, Excel, or any spreadsheet application.

Opt-in iCloud sync mirrors your usage records to iCloud so the dashboard shows combined data across your Mac and iPhone or iPad. Enable it in Settings → Usage → Sync usage via iCloud. Sync is off by default.

Request costs and token counts may be estimated. Provider usage is shown when returned. Direct provider prices do not establish Setapp credit charges. Check provider or Setapp account records for authoritative charges.

The menu bar widget adds a small item to the macOS status bar (top-right of the screen) that shows your running estimated spend at a glance without opening the main app. A typical title looks like ~$0.42 · ~38c, where the dollar figure is your estimated BYOK spend today and the credit count is your estimated Setapp usage today.

The widget is off by default. Enable it in Settings → Appearance → Show in menu bar. The widget is macOS-only and does not appear on iOS.

Clicking the menu bar item opens a compact popover with:

• Total tiles for today, this week, and this month
• Per-provider tiles showing each configured provider's usage
• Live balances for OpenRouter and DeepSeek (if configured), with a 60-second cache and a manual refresh button
• On the Setapp Mac edition, available account balances are fetched through the Setapp SDK. If a refresh fails, the app reports that state and may retain the last known balance. Account balance is not a receipt for one request.
• An Open Dashboard button that takes you directly to the full Usage Dashboard in Settings

A brief footer at the bottom reads: Estimates · open dashboard for detail.

For OpenRouter and DeepSeek, AI Council periodically fetches your remaining balance directly from the provider's API. Results are cached for 60 seconds. You can force an immediate refresh by tapping the refresh button in the Usage Dashboard or the menu bar popover.

Balances are account snapshots and may be cached. They are separate from request estimates. Provider support varies; an unavailable balance is not a zero balance.

On the Setapp Mac edition, available account balances are fetched through the Setapp SDK. If a refresh fails, the app reports that state and may retain the last known balance. Account balance is not a receipt for one request.

Request costs and token counts may be estimated. Provider usage is shown when returned. Direct provider prices do not establish Setapp credit charges. Check provider or Setapp account records for authoritative charges.

Setapp website

On Setapp builds, the Add Model picker shows two parallel sections for any provider that is available via Setapp:

• Setapp: <Provider> - badged via Setapp credits. Choosing a model here routes it through the Setapp gateway and consumes your Setapp AI credits.
• <Provider> - badged via your key. Choosing a model here routes it directly using your own API key for that provider.

The Setapp sections appear at the top of the picker regardless of whether you also have a BYOK key configured for the same provider.

Review the named route before confirming. An approved review uses that route and does not silently switch providers after a failure.

A model explicitly selected via Setapp uses the Setapp gateway. A direct provider route uses that provider account. Inspect the route in the review preview.

Yes. You can add models from both sections in the same council. For example, you might add a GPT model from Setapp: OpenAI (consuming Setapp credits) and Claude Opus from Anthropic (consuming your Anthropic BYOK key) in the same deliberation. Each model follows the route it was added under.

Request costs and token counts may be estimated. Provider usage is shown when returned. Direct provider prices do not establish Setapp credit charges. Check provider or Setapp account records for authoritative charges.