Prepended to every prompt in sessions belonging to this project.
Files uploaded here are injected into every prompt in this project's sessions. Any file type (10MB cap).
Move "" to:
Connected
Showing all messages
KoraCouncil
Pick how the council works on your question.
Session files:
×
Choose Modetap to switch below
── Council Modes ──
Council Debate
Consensus — One Answer
Brainstorm — Idea Generation
Research — Deep Dive
Verify — Check Work
Clarify — Ask First
── Executive Modes ──
Assignment — Plan + Execute
Brief — Exec Assistant
Strategize — Strategic Analysis
Review — Life/Biz Org
Coordinate — Daily Rhythm
── Operations Modes ──
Communicate — Draft Message
Concierge — Arrange It
Monitor — Project Watch
── Agentic Loops (gated) ──
Coding — Build (gated)
Task — Do It (gated)
Super Coding — Orchestrated (gated)
Agent — Server Tools (gated)
── Create ──
Image — Generate
Prompt Builder — Craft a Prompt
── All Models ──
All Models
Claude
Claude — Expert
Claude — Assistant
Claude — Intern
ChatGPT
ChatGPT — Expert
ChatGPT — Assistant
ChatGPT — Intern
Gemini
Gemini — Expert
Gemini — Assistant
Gemini — Intern
Grok (xAI)
Grok — Multi-Agent
Grok — Reasoning
Grok — Intern
DeepSeek
DeepSeek — Expert
DeepSeek — Assistant
DeepSeek — Intern
Qwen
Qwen — Expert
Qwen — Assistant
Qwen — Intern
MiniMax
MiniMax — Expert
MiniMax — Assistant
MiniMax — Intern
GLM
GLM — Expert
GLM — Assistant
GLM — Intern
Kimi
Kimi — Expert
Head Coder
Settings
General
App-wide preferences. Council member assignments + arbiter + coder live in the Council pane (admin only).
Which mode to use when you first open the app.
Prefer responses in a specific language (leave blank for English).
Adjust font and icon sizes across the entire interface.
AA100%
Tools & Connectors
Everything your seats can do — web search, reading URLs, files, and connectors — in one place, with a live map of which seat can do what.
Who can do what
Loading capabilities…
Web Search & URLs
Let seats fetch live web results + read links a user provides. On by default; uncheck to disable per provider. Default tools are available on demand — models call them only when a question needs them.
Supported file types
Attach these in chat — the + menu or drag & drop:
Connectors & tools
My ConnectionsConnect Google Drive, Gmail, Calendar & more (per-user OAuth).
Agent / Server AccessYour encrypted SSH + GitHub credentials for Agent mode.
MCP Servers & Tool CatalogRegister MCP tool servers + browse the installed tool catalog (admin; one shared store).
Personalization
Customize how the Council responds to you.
The more you put here, the more the council gets to know you. All fields optional. Auto-injected into every prompt — no more "what's your name?" intros.
Upload a profile image (shown in the sidebar instead of your initials). Square crop, resized to 256px. JPG / PNG / WebP.
The council's behind-the-scenes narration ("Research now convenes the full council…", stage explainers). Hidden by default to keep chats clean — errors, failures, and approval prompts always show regardless.
Prepended to every prompt — set tone, formatting, behavior preferences. Stored server-side so it works on every device.
Persistent notes about you that the council reads on every prompt — across all sessions. Council members can also append to this when they learn durable facts about you. Edit freely.
Stacked = traditional vertical. Columns = council member responses side-by-side, truncated with "Show full" toggle (handy when you want to compare answers).
Get notified when a council round completes or the arbiter finishes — even when this tab is closed (works on iOS Safari 16.4+ as a Home Screen app).
LLM Models
Choose which models to use for each tier. Use Refresh to pull latest models from each provider.
Claude / AnthropicCLI
Via OpenRouter: native web search, tools & prompt caching are disabled for this seat.
ChatGPT / OpenAICLI
Via OpenRouter: native web search, tools & prompt caching are disabled for this seat.
Gemini / GoogleCLI
Via OpenRouter: native web search, tools & prompt caching are disabled for this seat.
Grok / xAIAPI
Via OpenRouter: native web search, tools & prompt caching are disabled for this seat.
DeepSeekCLI
QQwen / AlibabaAPI
MMiniMaxAPI
GLM / Z.aiAPI
Kimi / MoonshotAPI
Prompt Builder AI
Default model that powers the Prompt Builder interview + generation. This is set by admins here and applies to all Prompt Builder sessions. Pick a provider default / CLI seat / custom seat — or ANY concrete model: type to filter, "↻ Load all models" pulls every live model from all providers + the OpenRouter catalog. Free text works too (a bare model id or an OpenRouter slug like z-ai/glm-5.2).
Model Pricing ($ / 1M tokens)▸
Cost shown under each response is computed from this table (keyed by the resolved model ID). Manual edit is the source of truth — almost no provider exposes live pricing via API, so "Re-seed defaults" only re-applies the researched seed values for you to review.
The CLI coder has web access — it looks up current published rates for the model IDs above and shows you a diff to review. A full-table lookup can take a few minutes. Nothing is written until you click Save Model Settings.
Chat auto-title
The cheap background model that names a new chat from its first message (so sessions don’t stay “New chat”). Runs once per session; never overwrites a manual rename. Empty = a fast Gemini default.
Saves settings AND tells every connected user to reload their dropdown labels immediately. Settings are global — every user sees the same model assignments.
Data & Files
Manage global context files injected into every prompt.
These files are injected as context into ALL prompts for ALL sessions.
Account
Manage your account and security.
My Connections
Connect your own accounts so the council can act on your behalf — read your Drive, draft your Gmail, check your Calendar. Each connection is private to you and your tokens are encrypted at rest. Disconnect any time.
Loading…
Loading…
Task Blueprints
Review and edit a task blueprint before approving it for execution. Each blueprint is private to you. Expand one to edit its objective, summary, constraints and deliverables, then Approve when it's ready to run.
Loading…
Audit Log
A timeline of every external action taken on your behalf — what was done, who/which agent did it, the tools and sources used, and its approval state. Newest first.
Showing the most recent events.
Loading…
Task → Model Routing
Pick which model handles each task type. When you send a single-model query (not a Council debate), the system auto-detects the task type from your prompt and routes it to the model you assign here. This is model selection, not role assignment — council member roles live in Settings → Roles. Leave a type blank to skip routing for it.
Code Generationwriting new code
Code Reviewbugs, security, quality
Planning & Architecturesystem design, roadmaps
Creative Writingstories, copy, marketing
Data AnalysisCSV, charts, SQL
Summarizationcondense documents
Math & Reasoninglogic, proofs, calculations
Researchfact-checking, deep analysis
Vision / Image Analysisunderstand uploaded images
General Chatconversation, Q&A
Pipeline (General)
Pipeline-wide settings that apply across modes, not just one council.
When the V2 (efficient) pipeline is active, this model condenses the prompt + history into the brief that the council reads. Pipeline-wide.
How many full debate rounds run automatically before the council pauses for your decision (another round vs. send to the arbiter). Default 1 = pause after Round 1. Set to 2 to auto-run two rounds before pausing. Debate mode only.
In a multi-turn session the council's running summarizer eventually folds your first message into a paraphrase. When ON (default), your original request is kept verbatim at the top of the shared context for every mode (debate, consensus, brainstorm, research, assignment, and all the rest) so no mode ever answers blind to your exact original ask. The composer's Full Prompt toggle forces this per-send even when this is off.
API Keys
Configure API keys and connection modes. Connection Mode sets how each provider is invoked as a council seat (Debate, Consensus, Brainstorm, Research, Chat). Auto = API if key exists, else CLI. (The autonomous Coding & Task loops route separately under Execution & Roles.)
ChatGPT / OpenAI
Ensure CLI is logged in. Need to re-auth?
Gemini / Google
Ensure CLI is logged in. Need to re-auth?
Claude / Anthropic
Ensure CLI is logged in. Need to re-auth?
Grok / xAI
Grok always uses API. Key is pre-configured.
DeepSeek
CLI mode uses opencode for direct file system access. DeepSeek participates in council debates only when a key is set.
QQwen / Alibaba
DashScope International (Singapore) endpoint. Qwen joins council debates only when a key is set.
MMiniMax
MiniMax International (minimax.io). MiniMax joins council debates only when a key is set.
GLM (z.AI)
GLM (glm-5.2) now calls z.AI directly (api.z.ai) using your own z.AI API key. Paste the key you generated at z.ai — it's required for GLM to join council debates.
Kimi (Moonshot)
Kimi (kimi-k2.7-code) now calls Moonshot directly (api.moonshot.ai) using your own Moonshot API key. Paste the key you generated at platform.moonshot.ai — it's required for Kimi to join council debates.
OpenRouter
One key, every model. Flip any provider to OpenRouter routing in Settings → LLM Models and its council seat runs through openrouter.ai on this key — no direct account needed for that provider.
Council
Every dropdown below is populated live from each provider's API. Click Refresh All to repull the latest model lists. The "fallback" runs when a council member errors out, so the debate doesn't degrade.
Claude seat
ChatGPT seat
Gemini seat
Grok seat
DeepSeek seatkey required
Qwen seatkey required
MiniMax seatkey required
GLM seatkey required
Kimi seatkey required
Add ANY OpenRouter model as an extra council seat — it joins the roster and the Roles pane automatically. Needs the OpenRouter API key (Settings → API Keys). Custom seats always run through openrouter.ai.
Build a named council from ANY seats — e.g. a two-model debate. Each seat picks its provider AND its model (type to filter the live catalog, or paste any id) plus an optional role — and the SAME provider can hold several seats on different models (Gemini Pro vs Gemini Flash, Opus vs Sonnet). Everyone picks a council per chat from the 🏛️ Council chip under the message box (council modes only).
When a council member errors / times out / returns empty, that seat is retried with this model. Pick something cheap and reliable.
MCP Servers — Advanced server registration
Register remote MCP servers by raw URL + auth. Each server can be allowlisted per council member. Wired into Claude (via anthropic-beta: mcp-client-2025-11-20), OpenAI Responses API, and Grok Agent Tools. Examples: filesystem, github, gmail, slack, playwright, custom. For a curated, one-tap install list, use Tool Catalog instead — both write to the same server store, so don't register the same server in both.
⚠ Safety: MCP servers can take real-world actions (send email, run code, modify files). Only register servers you trust. Auth tokens are stored server-side and never exposed in the UI.
Prompt Library — Curated expert prompts
Edit the curated prompt library users browse from the Prompt Builder. Each prompt is filed under a mode-group category and can carry a suggested mode so "Send to mode" starts the right session. Add, edit, or remove prompts here — no code change needed. Saves to data/prompt_library.json (a dated backup is written before every save).
Role-Based Settings
The one place for role-based settings — pick a mode, set the role assignments. Every role-capable mode (Debate, Brainstorm, Consensus, Assignment, Coding, Task, Research, Strategize, Verify, Brief, Monitor) is edited here: choose a mode, edit each role's name and one-line mandate, and assign which provider plays each role. Saves to config.roles.<mode> and takes effect on the next run of that mode.
Loading…
Available agent personas
Reference only. The Coding & Task loops assign work by provider (above); each provider runs with the built-in role prompt for its assigned seat. These are the canonical personas from /opt/agent-rules/personas.
Execution
Controls how the autonomous loop modes (Coding & Task) execute: CLI vs API per loop, and the runaway caps. CLI drives the provider's local agent (best for coding/task); API calls the hosted model directly. These are workspace defaults.
Council modes (Debate, Consensus, Brainstorm, Research, Chat) do not use this map — their CLI/API routing is set per provider in LLM Connections → Connection Mode.
Role packs for every mode (including Coding & Task) are edited in Settings → Roles.
Loop-mode execution
Pick CLI or API for the Coding and Task loops. Saves to layerb.mode_execution. Only these loop modes read this — council modes route per provider (above).
Loop caps
Hard ceilings for autonomous loops — protect against runaway iterations and spend.
Super Coding
Head-coder orchestration: your chosen lead LLM delegates subtasks to best-fit models in parallel waves. These caps are hard ceilings (spend is charged before each call).
Looking for the loop role assignments (which provider plays Architect / Implementer / Verifier …)? Those live in Settings → Roles — pick the Coding loop or Task loop mode there. The available agent personas are listed there too.
Tool Catalog — One-tap tool install
Add MCP tool servers in one tap from the curated catalog, or paste any server URL. Each installed server can be scoped to specific providers and modes. This shares the same server store as MCP Servers (Advanced) — don't add the same server in both places.
⚠ Safety: MCP servers can take real-world actions (send email, run code, modify files). Only add servers you trust. Auth tokens are stored server-side and never shown in the UI.
Catalog — one-tap add
Installed servers
User Management
Create and manage accounts. Three roles available: Admin (full access), Developer (chat + code modes), User (chat only).
KoraCouncil — Guide
What KoraCouncil is
KoraCouncil is a council of nine top AI models — Claude, ChatGPT, Gemini, Grok, DeepSeek, Qwen, MiniMax, GLM, and Kimi — working on your question together. Instead of trusting one model, you get the council: they debate, check each other's work, and hand you a single answer that's stronger than any one of them alone. You can always open up any individual member's response to see who said what.
And it's much more than debate. The same council runs in a whole range of modes — researching a topic, verifying a piece of work, brainstorming ideas, writing you an executive brief, drafting a message in your voice, arranging plans, watching your open projects, even generating images. With your permission, it can also act on your connected accounts (Drive, Gmail, Calendar). The Modes tab walks through each one.
The chat bar
Everything starts at the chat bar at the bottom — the same on web and on the iPhone app:
+ button (left) — opens a menu to attach files or scan a QR code. On the iPhone it also carries a small mode dot and is where you switch modes.
Choosing a mode — on the web, the mode pill in the bar (or ⌘K) opens the mode picker; on the iPhone, tap the + button or open the sidebar to switch modes.
Image mode — the ✦ button (web header) / the + menu → Create Image (iPhone) flips the chat into Image mode so you can generate pictures. See the Image Mode tab.
↑ send arrow (right) — sends your message. While the council is working, it becomes Interject (web) / a red Stop with Interject in the + menu (iPhone) so you can steer mid-run.
Smart placeholder — the prompt text inside the bar changes with the mode, so you always know what your next message will do.
A brand-new session opens on a landing that lays out every mode in columns, with Prompt Builder featured up top — tap a mode to pick it, or let Prompt Builder draft your prompt with you.
Your first question
Leave the mode on Council Debate (the default), or pick another from the mode pill.
Type your question and tap ↑.
The council works through a couple of rounds — you watch each member's answer stream in.
When they're done, you decide: ↻ Add Another Round for more refinement, or Send to Arbiter for the final synthesis.
After the Arbiter answers, keep adding rounds or just chat about the result — see the Discussion / Filter tab.
Getting started in three steps
Ask anything. Type your question in the chat bar and hit ↑. The default Council Debate mode is a great starting point for almost any question.
Pick the right mode when it matters. Every new session greets you with a mode-picker landing — columns of modes with Prompt Builder up top. You can also switch any time: on the web tap the mode pill in the chat bar (or press ⌘K / Ctrl-K); on the iPhone tap the + button or open the sidebar. Want a research deep-dive? A second opinion on some work? A drafted email? There's a mode for it — see the Modes tab.
Make it yours. Use Projects to group related chats with shared instructions and files, connect your accounts under Settings → My Connections, and add a note to User Memory so the council remembers your context across every session.
Finding your way around
Left sidebar — your current mode, Projects, and Sessions live here, plus a search box that finds any past chat by name or content. See Projects & Sessions.
Collapsed icon rail (web) — collapse the sidebar and it becomes a slim icon rail; the Search, Prompt Builder, Modes, Projects and Sessions icons open a flyout right there without re-expanding.
Mode picker (⌘K) — the fastest way to switch modes. Just start typing the name.
Settings — the gear at the bottom of the sidebar opens your connections, projects, personalization, blueprints, and audit log.
This guide — the Guide button up top reopens it any time.
Prompt Builder — it writes the prompt with you
Not sure how to ask? Prompt Builder interviews you and drafts a strong prompt. Open it from the accent Prompt Builder button on the new-session landing, the ✎ Prompt Builder icon on the collapsed sidebar rail, or the pinned Prompt Builder row at the top of the sidebar's mode list.
Pick a starting point — Research, Writing, Image, or "I need to explain & attach files" — then answer a few quick questions. It's never a dead end: one button reads Next while it's still interviewing you and Submit the moment it has enough (an accent ✦ Build prompt now line also appears so you can build any time).
When the prompt is ready you can Copy it, Save to my prompts, or send it straight into a mode: the builder suggests 3–4 modes that fit your prompt (an Image prompt always offers Image mode), and below the suggestions every mode is laid out as pills in the same category groups as the mode picker — tap one to start that session pre-filled with your prompt. Prompt Builder chats live in their own Prompt Builder group in the sidebar.
Voice notes — speak your prompt
Tap the 🎙 mic in the composer tool-rail, speak for as long as you like, then Stop. Your audio is transcribed and gently cleaned for reading — filler removed, punctuation and paragraphs added — without changing your words. Hit Use in Prompt Builder and the readable text drops straight into the composer so you can start building. Every note is saved to your Library (the mic panel’s Library tab): reopen any note to read it, play it back karaoke-style with the words highlighting as they’re spoken, rename or delete it, or feed it into the Prompt Builder again.
Prompt Library
Tap Browse library inside Prompt Builder to open the Prompt Library: curated expert prompts grouped by category, plus My prompts (anything you saved). Copy any prompt or Send it to a mode. You can also manage My prompts under Settings → Prompt Library.
Modes — pick the right tool for the job
Tap the mode pill in the chat bar, or press ⌘K / Ctrl-K, to open the picker. Start typing to filter. Modes are grouped — here's what each one is for.
Pin & ReorderThe sidebar's quick-mode rows are your pins. Open the mode picker (⌘K) and tap the pin icon on any mode to add it to — or remove it from — the sidebar. Drag a sidebar row up or down to reorder it (desktop; on touch, use the picker's pin toggles). Hover a sidebar row and hit × to unpin it. Your pins and their order sync to your account on every device. Prompt Builder always stays pinned at the top.
Council
Council DebateThe default. All members debate, critique each other, and synthesize one consolidated answer. Best all-rounder. Want each member arguing from a distinct hat? Flip the chat-bar Parallel | Role toggle to Role — that's the old "Role Debate", now built into Debate itself.
Reasoning EffortThe controls row under the chat bar (like the Claude app) holds the + button and the chips: Mode (tap to open the full mode picker), 🎭 council-style, 📝 prompt-handling, and the ⚡ effort chip showing the current level (Auto | Low | Med | High) — tap any chip for a small sheet where each option is explained. It sets how hard every model thinks for this session — change it any time, even mid-conversation (your next message applies it). Auto uses each model's admin default (Settings → LLM Models, per provider). Low = fastest & cheapest, High = deepest reasoning (slower, costs more). Models without an effort dial simply ignore it.
OpenRouter Routing(Admin) Settings → API Keys holds a single OpenRouter key, and every provider card in Settings → LLM Models has a Routing selector: Direct (default — the provider's own API + key) or OpenRouter (that seat runs through openrouter.ai on the shared key — handy when a provider's direct account is out of credit; the seat stays on the council without its own key). The OpenRouter model box sets the exact slug (e.g. z-ai/glm-5.2); leave it blank for the sensible default shown as the placeholder. Note: routed Claude / ChatGPT / Gemini / Grok seats lose native web search, tools & prompt caching.
Custom Seats(Admin) Settings → Council → Custom Seats: press + Add Seat, give the seat a name and an OpenRouter model slug (e.g. meta-llama/llama-4-maverick), and a brand-new council member joins the roster — no direct provider account needed, it runs through openrouter.ai on the shared OpenRouter key. Each custom seat shows up in the Roles pane for hat assignments automatically. Toggle Enabled to bench a seat without deleting it, or Remove to drop it.
Custom CouncilsBuild your own council. Want a debate between just TWO models? (Admin) Settings → Council → Custom Councils → + Build Council: name it, then add seats — each seat is a provider + a model (tap the model box to browse the live catalog, type to filter, or paste any id; leave blank for the seat default) + an optional role (architect, critic, researcher…). The same provider can sit on several seats with different models — pit Gemini Pro against Gemini Flash, or Opus against Sonnet. Then anyone can pick it per chat from the 🏛️ Council chip under the message box (shown in council modes — Debate, Consensus, Brainstorm, Research): choose Full Council or any custom council, and every send + follow-up in that chat uses exactly those seats. Councils with roles automatically run role-style. Edit a council any time; toggle Enabled to bench it, or Remove to drop it.
ConsensusThe council talks it through, agrees, and delivers one answer the whole council stands behind.
BrainstormWide-open idea generation across the council, then the best ideas get pulled together and ranked.
ResearchA two-stage deep dive: first a grounded web-retrieval pass (Gemini) gathers live sources, then the full council cross-checks and reconciles them into a Research Synthesis — with each member's card and rounds shown.
VerifyA second opinion on a piece of work (text, a plan, or code) — checks it for correctness and consistency against your request.
ClarifySurfaces the questions worth answering before acting, instead of guessing at what you meant.
Executive
AssignmentGoal-driven: a planner breaks your goal into a plan you approve, then the council executes it. Review and approve under Settings → Task Blueprints.
BriefA decision-ready briefing on your request, the way a great chief-of-staff would prep it.
StrategizeStrategic analysis — the situation, the opportunities, the risks, and the moves worth making.
ReviewReviews and organizes across the things on your plate — life and business.
CoordinateYour daily rhythm — operational coordination across everything you have open.
Operations
CommunicateDrafts a message in your voice for a channel (email, etc.). With your approval, it can send through your connected Gmail.
ConciergeResearches and arranges options against your preferences and budget, then lays the choices out for you.
MonitorWatches your open projects for stalls and things that need your attention.
Agentic Loops
CodingA guided build loop: a plan is set with a clear "done" target, the work gets implemented, and a different model verifies it — all within safe limits.
Super CodingPick a Head Coder — it splits your goal into subtasks and delegates each to the best-fit model (cheap models for simple parts, experts for hard parts, the Gemini managed agent for deep research), runs them in parallel waves, integrates everything, and a different model verifies. Your Head Coder can also hand a subtask to a CLI coding agent (Claude Code, Codex or Antigravity) that works in its own isolated copy of a repo — editing, building and running tests for real — then reports back and asks before anything is merged. Token and spend ceilings are off by default so a run always finishes; every response still shows its own token count and cost.
TaskA general do-it loop: plan → execute → verify. Anything real-world or irreversible stops at an approval gate first.
Find these by name in the mode picker (⌘K).
Create
ImageGenerate an image from a description, then keep refining in the same session. See the Image Mode tab.
Direct models (no council)
All ModelsAsk every model at once — each answers you independently, no debate or synthesis. Pick Best / Balanced / Fast on the composer toggle. Great for comparing answers side by side.
A single modelPick one provider at one tier for a straight one-on-one chat — no council, just that model.
Modes that take real actions on your behalf always draft a blueprint you approve first, and every external action lands in your Audit Log — see the Blueprints & Audit tab.
The Parallel / Role toggle
For council modes that involve several members, the chat bar shows a small Parallel | Role toggle. It changes how the council works on this run:
Parallel — every member works the question independently, as themselves. You get a spread of genuinely different takes.
Role — each member puts on a different hat (Architect, Skeptic, Researcher, Pragmatist, Visionary, Verifier, User Advocate…) and argues from that lens, then it's all synthesized across roles. Great when you want pressure-tested, well-rounded coverage.
The toggle applies to: Debate, Consensus, Brainstorm, Research, Verify, Brief, Monitor, Strategize, and Assignment. It only shows up when the current mode supports it, and it remembers your choice per mode.
Tip: try the same question once in Parallel and once in Role — they often surface very different things.
How rounds work
In a debate, the council runs a couple of rounds on its own — you just watch the answers come in. When they're done, a card appears asking what you'd like next: Send to Arbiter for the final synthesis, or Add Another Round for more back-and-forth. There's no limit — keep adding rounds until you're happy.
On iPhone the decision card reads "Round complete — your call" with Add a Round and a final-stage button (Send to Arbiter / Final Synthesis / Reach Consensus, depending on the mode).
Steer the next round
Type a note in the chat bar before you click Add Another Round, and that becomes your instruction for the next round ("focus on cost", "push back harder", etc.). Leave it empty for a general refinement.
Interject any time
You don't have to wait. Drop a message into the chat bar while the council is working — on the web the send button itself turns into Interject. Your note gets queued and the council picks it up at the next step to either steer the discussion or add it as context. You can interject as many times as you like, and each one is saved to the transcript so you can scroll back.
On iPhone, while the council is running the send arrow becomes a red Stop; to steer mid-run use the + menu → Interject.
Pick up where you left off
Refresh, close the tab, come back tomorrow — your decision card and round controls are exactly where you left them.
And if a run ever gets interrupted (a lost connection, a restart, a closed laptop) — in any mode, from a debate to a brainstorm to the Prompt Builder mid-interview — reopening the session shows a "Session paused — your work is SAFE" notice with a Continue button. One tap finishes the run from what was already saved: completed answers are kept, only the missing pieces run, and you never retype a thing.
When a member fumbles
If a member's response comes back broken or it doesn't finish, the council flags it and tucks it behind a one-line "Response flagged" shade (tap to peek). You decide what to do with that seat:
Re-run — have that member try again.
Reconsider — ask them to take another pass at it.
Keep — keep the flagged answer anyway.
Drop — leave that member out of this round.
Nothing finalizes while a fumble is waiting on you — you stay in control of the result.
After the debate — keep talking
Once a debate finishes, the bar at the top of the chat gives you filter buttons (All / Claude / ChatGPT / Gemini / Grok / DeepSeek / Qwen / MiniMax / GLM / Kimi) alongside the round controls. You can keep chatting about the result right there.
Filter the transcript
Tap a member's name and only their messages stay on screen — your own prompts always stay visible. Tap All to bring everyone back. These filters are just a view; they don't change who answers next.
Talk to one member
To direct your next prompt at a single model, pick that model from the mode picker (e.g. "Claude — Expert"). Your message goes only to them — a one-on-one.
@mentions
Start a prompt with @claude, @chatgpt, @gemini, @grok, @deepseek, @qwen, @minimax, @glm, or @kimi to address that member directly.
Image mode
Tap the ✦ sparkle button in the chat bar — the chat flips into Image mode (the mode pill reads "Image — Generate"). From there, just type a description and hit ↑ to generate. Keep typing new prompts in the same session to keep iterating. You can also pick Image from the mode picker (⌘K).
Refine an image
Every image you make has a Use as reference / Refine button. Tap it and that image loads back in as a reference with your prompt seeded — perfect for "same idea, but…" tweaks.
Reference images
You can also feed in your own reference images and choose the model, aspect ratio, count, and style from the image controls. Generated images stay inline in the chat across refreshes — right-click → Save As to download.
Quick one-off: type /imagine your prompt from any chat to generate without switching modes. Switch to Image mode when you want to keep iterating.
The model picker
Tap the mode pill in the chat bar (or press ⌘K) to choose what happens next. Start typing to filter — modes and models both show up. For what each mode does, see the Modes tab.
Council, parallel, or one-on-one
Council Debate (default) — the full council works your question together.
All Models — ask every model at once; each replies independently, no synthesis. Pick Best / Balanced / Fast on the composer toggle. Great for a quick side-by-side.
A single model — pick one provider at one tier for a direct one-on-one.
Three tiers
Each provider comes in three tiers, so you can trade depth for speed and cost:
Expert — the strongest model, for the hard stuff.
Assistant — fast and capable, the everyday workhorse.
Intern — lightweight and quick, for simple fan-outs.
The nine members
The council is Claude, ChatGPT, Gemini, Grok, DeepSeek, Qwen, MiniMax, GLM, and Kimi. Here are the models currently behind a few of them:
Claude
…
ChatGPT
…
Gemini
…
Grok
…
DeepSeek
…
Qwen, MiniMax, GLM, and Kimi round out the council too. Use the picker's filter box to jump to any specific model.
Tools & Connectors — one place
Everything the council can reach lives under Settings → Tools & Connectors: web search, reading URLs, files, and your connected accounts. At the top is a simple “who can do what” grid — one glance shows which model can search the web, read a link, use files, or reach your connectors, and how (native, OpenRouter, or a tool). Web search is on by default; an admin can tune it there.
My Connections
Connect your own accounts so the council can work with your real stuff — read your Google Drive, draft your Gmail, check your Calendar. Open Settings → Tools & Connectors → My Connections and connect each one with a single tap.
Private to you — your connections are yours alone, encrypted, and never visible to anyone else's council.
Google Drive — the council can read and search your files for context.
Gmail — read for context, and draft (or, with your OK, send) messages in Communicate mode.
Google Calendar — read your schedule for coordination and concierge tasks.
Disconnect any time — revoking a connection cuts off access immediately.
MCP Servers & Tool Catalog
Want more? In Settings → Tools & Connectors, admins can open MCP Servers & Tool Catalog — register tool servers and browse the ready-to-add tools (one shared list). Most are one tap to add; some ask you to sign in or paste a key. Once added, the council can use them on your tasks — all under the same approve-first, audited flow.
Files & default tools
The Tools & Connectors pane also lists the file types you can attach (images, PDFs, Office docs, text, spreadsheets & more) — the same list drives the + attach menu and drag-and-drop. Built-in tools (web search, URL-reading) are on demand — models call them only when a question needs them.
Works everywhere
Your connections live with your account, so they're the same on the web app and the iPhone app — connect once, use everywhere.
Anything done through a connection runs through a blueprint you approve first and lands in your Audit Log — see the Blueprints & Audit tab.
Agent Mode — a model runs commands on your server
Pick Agent — Server Tools from the mode picker to let one model run real shell commands over SSH on your own server (git status, builds, tests, deploys) through a guarded run_shell tool. Describe the task, hit Send, and the model works step by step.
Ask vs Autonomy
Ask (default, safe) — you approve every command before it runs.
Autonomy — commands run without asking, except anything dangerous (rm -rf, force-push, deploys, DROP/TRUNCATE, curl|sh …), which always asks — even here.
Approvals & questions — detailed choice cards
When a command needs your OK, a card shows the exact command, the model's reason for running it, and your choices — each a full card with a one-line description: Approve once, Approve & remember this session, Approve & remember always, or Deny. You can also type an instruction back to the model. If the model needs a decision, it asks the same way — labelled options, each with a short description, plus a type-your-own box.
Server Access — SSH + GitHub
Open Settings → Agent / Server Access to store your credentials, encrypted at rest and never shown again:
SSH host, user, and private key — use a non-root, limited-privilege account so a misbehaving command is contained.
GitHub token (optional) — a fine-grained or classic PAT. When set, the agent can use gh and authenticated git on your server — GitHub writes (open a PR, push) still ask first.
Your credentials are private to you and used only by your own Agent runs. Dangerous commands always confirm, in every mode.
Task Blueprints — approve before it acts
When you use a mode that actually does something (Assignment, Communicate, Concierge, Coordinate, and the Task loop), the council first writes up a blueprint — a plan of exactly what it intends to do. Nothing happens until you say go.
Open Settings → Task Blueprints. Each one is private to you. Open a blueprint to read or tweak its objective, summary, constraints, and deliverables, then hit Approve when it's ready. No blueprint, no action.
Audit Log — a record of everything
Open Settings → Audit Log for a timeline of every external action taken on your behalf, newest first. Each entry shows what was done, which tools and connections it used, and its approval state. Tap Refresh to pull the latest.
Approve-first plus a full audit trail means you're always in control of what the council does for you.
Sessions
Every chat is a session. They're listed in the sidebar, grouped by when you last used them (Today, Yesterday, and so on). Tap + New Session to start fresh.
Search
Use the search box at the top of the sidebar to find any past chat — it matches both the session name and the content of the messages inside, so you can find that conversation even if you don't remember what you called it.
Session actions
Each session row has a ⋮ menu (and a pin):
Pin — tap the star to keep important sessions at the top of the list.
Rename — give it a clearer name (or use the /rename command). New chats name themselves from your first message; a manual rename always wins and sticks.
Move to project — file it under one of your projects (or back to top-level).
Delete — remove it (you'll be asked to confirm).
Projects (workspaces)
Projects group related sessions under a folder with their own shared instructions and files. Tap + New in the sidebar's Projects section to create one, then tap a project name to expand or collapse its sessions.
Project settings
Tap the gear next to any project to open its settings:
Name — rename it any time.
Custom instructions — added to every prompt in this project's chats.
Project files — upload files (up to 10MB each) that the council sees on every prompt in the project. You can select multiple files at once.
Use global context toggle — keep your global instructions, memory, and files in the mix, or turn them off so the project runs purely on its own context. (Your User Memory always applies.)
Delete — the project's sessions move back to top-level (they're not deleted); the project's files are removed.
Super Coding sessions
The Head Coder picker beside the composer decides which model plans and integrates the run — change it any time, per session. Alongside the usual models you can pick an agentic CLI seat: slower to start, but it can read the actual repo before planning, which makes for a much better breakdown on a real codebase.
Super Coding sessions get their own group in the sidebar, sitting just above Sessions, so a long-running build never gets buried among your everyday chats. Tap + New there to start one — it opens a fresh session already set to Super Coding, ready for you to describe what to build.
Once you have a few, you can tidy them into folders: tap + New folder at the bottom of the group to make one, then use the little folder button on any session row to file it away. Folders rename, collapse, and delete from their ⋮ menu — deleting a folder only removes the folder, never the sessions inside it (those move back to the ungrouped list). Folders are remembered in this browser.
A Super Coding session always lands in this group — no matter which project you started it from. If you began it inside a project it stays filed under that project too; the group is a lens on everything Super Coding, the project is where it lives. Prompt Builder works the same way.
Your folders follow you between devices. File a session on your laptop and it is filed on your phone and in every other browser too. Whether the group is open or collapsed stays local to each device on purpose — collapsing it on your phone will not collapse it on your desktop.
Web access
The council can read the live web. Two ways to trigger it:
1. Paste a URL
Drop any https://... link into your prompt. Up to 5 URLs per message. Example:
Check this page and tell me what you think:
https://example.com/article
2. Ask them to search
Use phrases like "search the web for X", "google X", "look up X", "find online X", or "web search for X". Example:
Search the web for the latest Pokémon TCG set release dates
Give the council files to read
The council can read documents and images you share. Attach from the + menu, drag and drop onto the chat, or paste straight in. Supported types include:
Images — PNG, JPG, and friends (the council can see them).
Documents — PDF, Word (.docx), Excel (.xlsx), PowerPoint (.pptx), CSV, Markdown, plain text, and more.
Where files can live
In a message — attach a file to a single prompt.
In a project — project files ride along with every prompt in that project (see Projects & Sessions).
Globally — Settings → Personalization lets you set global files and custom instructions that apply everywhere.
Global + session memory
Open Settings → Personalization → User Memory to jot down durable facts about you — your role, your preferences, what you're working on. Your User Memory and global instructions ride along on every prompt in every mode, across all your sessions, and the council can add durable facts over time. It's yours and private.
Canvas
The Canvas button (in the header) pulls code blocks out of the council's answers into a side panel where you can save, copy, or diff them.
Chat layout
Under Settings → Personalization → Chat Layout, choose how council answers are shown:
Stacked (default) — a clean top-to-bottom flow of messages.
Columns — each round's member answers sit side by side as cards (a few across on desktop, fewer on tablet, one on phone). Each card has a Show full toggle to expand it.
Make it readable
Personalization also has a font-size control, so you can scale the text up or down to taste.
The sidebar
+ New Session — start a fresh chat.
Search — find any past chat by name or content.
Projects — your workspace folders.
Sessions — your chats, grouped by recency, pinned ones up top.
Your avatar + the settings gear at the bottom.
Collapse to an icon rail — collapse the sidebar and it becomes a slim rail; each icon (Search, Prompt Builder, Modes, Projects, Sessions) opens its section in a flyout without re-expanding.
Slash commands
/imagine [prompt]Generate an image without leaving your current mode.
/imagine gemini [prompt]Generate an image with Gemini instead of the default.
/rename [new name]Rename the current session (same as the rename action in the ⋮ menu).
@ mentions
@claude / @chatgpt / @gemini / @grok / @deepseek / @qwen / @minimax / @glm / @kimiAddress one member directly — they reply, the others stay quiet.
Keyboard
⌘K / Ctrl-KOpen the mode picker.
EnterSend your message.
Shift + EnterAdd a new line.
Export a session
Open the export option from the chat bar to download the current session. You can filter by date range and by who said it (everyone / AI only / you only), then save as:
PDF — a clean, color-coded transcript with the noise stripped out.
JSON — the full raw session, handy if you want the data.
Saving images
Generated images stay inline in the chat across refreshes — right-click → Save As to download any of them.
Native iOS app
There's a native iPhone app (currently via TestFlight). It uses the same iOS-style composer as the web (+ menu, ↑ send, mode-aware placeholder) and shares your account: your settings, connections, projects, and sessions sync across web and app. Settings are fully editable in-app on iOS too — including My Connections, Personalization, and push prefs — and any change persists back to the web.
Install as a web app (PWA)
iPhone Safari: tap Share → Add to Home Screen. Opens fullscreen with no browser chrome — looks and feels like a native app.
Android Chrome: menu → Install app, or Chrome will offer an Install banner.
Desktop Chrome / Edge: address bar → install icon (or menu → Install KoraCouncil).
Push notifications
Turn these on under Settings → Personalization → Push Notifications to get a nudge when:
A round finishes and a decision is waiting for you.
The Arbiter wraps up the final answer.
You can opt out of individual events. iOS note: push only works once you've added the app to your Home Screen (regular Safari can't receive it), on iOS 16.4 or later.
Your account
Your avatar and the settings gear sit at the bottom of the sidebar. From Settings you can:
Manage your connections and the Tool Catalog (Drive, Gmail, Calendar, and more).
Review and approve your Task Blueprints, and check your Audit Log.
Edit your User Memory and global instructions under Personalization.
Set your chat layout, font size, and push notification preferences.
Change your password.
Log out (the button at the bottom of the settings menu).
Same account, everywhere
Everything here — your settings, connections, projects, and sessions — syncs between the web app and the iPhone app, and you can edit it from either one.
Tips to get more out of the council
Start with Debate, switch when you have a reason. It's a great default. Reach for Research, Verify, Brief, or the others when the job is specific.
New here? Let Prompt Builder draft it. Hit Prompt Builder on the landing and let it write the prompt with you.
Save prompts you reuse. Add them to My prompts (Prompt Library), then Send them straight to a mode.
Learn ⌘K. The mode picker is the fastest thing in the app — just press ⌘K and type.
Try Parallel vs Role on the same question. Independent takes and role-based hats surface different things.
Add another round when the answer's close but not quite there — and type a note first to point the council at what to fix.
Interject instead of starting over. Drop a correction mid-run and the council folds it in.
Feed it your files. Attach a document or image and ask about it directly.
Use Projects for ongoing work. Shared instructions and files mean you don't repeat context every time.
Fill in User Memory once. A few facts about you make every answer more relevant.
Connect your accounts to let the council draft email, read a doc, or check your calendar — always with your approval first.
Lost a chat? Search. The sidebar search finds it by name or by what was said inside.