Got an EPIC feature idea Poppy should have?

PLEASE copy this template and click “New Feature Request“ 👇

The feature I think Poppy should have is…

This would help me because…

We promise the team and us will read every single idea and feedback.

- Naz & Qazi

Add "Boardroom" Button - Multi-Model AI Brainstorming Feature

Feature Spec: Boardroom Mode Multi-Model Collaborative Analysis with User-Selected Presenter Executive Summary Problem: Users currently get single-perspective outputs. Even when they try multiple models manually, there's no synthesis—just fragmented responses. Solution: Add a "Boardroom" action button that orchestrates 3 diverse LLMs through 5 specialized analysis steps, then synthesizes results through the user's selected model. Value Prop: Higher quality outputs (multi-perspective → fewer blind spots) Reduced iteration cycles (users don't manually run 5-10 prompts) Differentiated feature (no competitor offers orchestrated multi-model analysis) Leverages existing model infrastructure (no new APIs needed) Technical Feasibility: High. Uses existing model integrations + lightweight orchestration layer. User Experience Flow Before (current state): User types prompt Selects model from dropdown Gets single-model response If unsatisfied, manually tries different models/rewrites prompt After (with Boardroom): User types prompt Clicks Boardroom button (bottom action row) Waits 8-15 seconds Gets synthesized response from their selected Presenter model Optional: Expands "Show Boardroom Notes" to see analysis breakdown Key UX principle: Zero configuration. User doesn't pick council models or configure steps—just click and get better output. Technical Architecture Component Overview [User Prompt] ↓ [Boardroom Orchestrator] ↓ [Parallel Multi-Model Execution] (Steps 1-5) ↓ [Bundle Aggregator] ↓ [Presenter Synthesis] (User's selected model) ↓ [Final Output to Chat] Execution Flow (Detailed) Phase 1: Council Execution (Parallel) Input: User's prompt Process: Send user prompt + role-specific instruction to 3 pre-selected models Run 5 sequential steps (each step parallelizes across 3 models) Each model returns max 5 bullets per step Output: 15 structured responses (3 models × 5 steps) Phase 2: Bundle Creation Input: 15 council responses Process: Organize by step into structured bundle Output: Markdown-formatted analysis bundle Phase 3: Presenter Synthesis Input: Original prompt + bundle Process: Selected model (from existing dropdown) writes final response Output: Clean, conversational answer in chat The 5 Analysis Steps Each step is a specialized lens. All 3 council models run each step. Step Role Output Focus 1 Strategist Best approach, structure, execution plan 2 Creative Hooks, angles, format variations 3 Editor Weaknesses, gaps, what to cut 4 Market/Viral What performs, packaging guidance 5 Bias Detector Creator vs. audience mismatches Why 5 steps: Covers strategic, creative, editorial, market, and reality-check lenses. Universal across use cases (content, brand, strategy, copy). Council Model Selection Potential Default: Model A: GPT-4o-mini (OpenAI) Model B: Claude 3.5 Haiku (Anthropic) Model C: Gemini 1.5 Flash (Google) Why these: Fast (low latency) Cheap (cost-effective for 15 calls) Diverse families (different training, strengths, biases) Alternative: Dynamic selection based on availability/cost, but must ensure 3 different model families. Presenter Model Selection Behavior: Whatever model user has selected in the existing dropdown becomes the Presenter. Why this works: User gets output in their preferred model's voice/quality Separates "background compute" from "presentation layer" No new UI needed—leverages existing dropdown Example: Council runs on: GPT-4o-mini, Haiku, Flash User has selected: Claude Opus Final output: Written by Opus (synthesis of council work) Token/Cost Analysis (Hypothetical) Per Boardroom Execution: Council Phase: 15 calls (3 models × 5 steps) ~300 tokens input per call (prompt + role instruction) ~200 tokens output per call (5 bullets) Total council: ~7,500 tokens Presenter Phase: 1 call ~4,000 tokens input (original prompt + bundle) ~800 tokens output (final answer) Total presenter: ~4,800 tokens Grand Total: ~12,300 tokens per Boardroom execution Cost Comparison: Council models (mini/haiku/flash): ~$0.015 per execution Presenter model (varies): ~$0.02-0.15 depending on model Typical total: $0.03-0.17 per Boardroom vs. Manual Approach: User runs 5-10 separate prompts trying to get good output Higher aggregate token usage More user time burned ROI: Higher per-click cost, but dramatically fewer clicks needed. Integration with Existing Features Search/Connected Sources: Behavior: If user has Search enabled or board sources connected, council models inherit that context automatically. Implementation: No special handling needed—just pass context to council models same way you pass to single-model calls. Model Dropdown: Behavior: Dropdown becomes "Presenter selection" when Boardroom is used. Implementation: No UI change needed—just change backend behavior: selected model does synthesis instead of primary generation. Implementation Phases Phase 1: MVP (Ship This First) ✅ Boardroom button (bottom action row) ✅ 3 fixed council models (mini/haiku/flash) ✅ 5-step orchestration ✅ Presenter synthesis ✅ Final output to chat (no notes visibility) Scope: ~2-3 weeks, 1 backend engineer + 1 frontend engineer Phase 2: Polish "Show Boardroom Notes" toggle (collapsed by default) Display step-by-step breakdown Scope: +1 week Phase 3: Advanced (Optional) Fast/Deep toggle (2 steps vs 5 steps) Custom council model selection (power users) Boardroom analytics (track quality improvement) Success Metrics Adoption: % of active users who try Boardroom in first 30 days Repeat usage rate (users who use it 3+ times) Quality: Reduction in follow-up prompts after Boardroom vs. standard User satisfaction scores (survey) Retention increase among Boardroom users Efficiency: Avg time-to-acceptable-output (Boardroom vs. manual iteration) Risk Mitigation Risk Mitigation Latency (8-15s feels slow) Progress indicator, set expectations ("Consulting 3 models...") Token costs spike Hard limits on council output (5 bullets), monitor usage patterns Users don't understand it Clear onboarding tooltip, example use cases Presenter model doesn't synthesize well Pre-test synthesis prompts across models, fallback to structured format Why This Feature Wins Differentiated: No other AI chat tool offers orchestrated multi-model analysis Leverages existing infra: Uses models you already have integrated High perceived value: "3 AI models working together" = premium feel Reduces churn: Users get better outputs → stay longer Upsell opportunity: Can gate advanced features (custom councils, more steps) Proposed System Prompts Council Role Templates Step 1: Strategist You are the Strategist in a 3-model Boardroom. User's request: {USER_PROMPT} Your job: Identify the best approach, structure, and execution plan. Return (5 bullets max): - Primary goal and success criteria - Recommended structure/framework - Critical first steps - Key dependencies or requirements - One thing most people miss when doing this Step 2: Creative You are the Creative in a 3-model Boardroom. User's request: {USER_PROMPT} Your job: Generate hooks, angles, and format variations. Return (5 bullets max): - Strongest hook/opening angle - 2-3 alternative approaches - Format recommendation - One unexpected creative angle - What makes this stand out Step 3: Editor You are the Editor in a 3-model Boardroom. User's request: {USER_PROMPT} Your job: Find weaknesses, gaps, and what should be cut. Return (5 bullets max): - Biggest weakness in this approach - Critical missing information - What to remove/simplify - Assumptions that need validation - One question that must be answered first Step 4: Market/Viral Lens You are the Market Analyst in a 3-model Boardroom. User's request: {USER_PROMPT} Your job: Assess what performs and how to package for maximum impact. Return (5 bullets max): - What format/style wins right now - Audience hook priority - Pacing/delivery guidance - What's likely to underperform - Packaging recommendation Step 5: Bias Detector You are the Bias Detector in a 3-model Boardroom. User's request: {USER_PROMPT} Your job: Call out creator bias vs. audience reality. Return (5 bullets max): - Where creator preferences conflict with audience needs - Self-confirmation bias flags - "You'd like this but your audience won't" moments - Overcomplexity warnings - Reality check: what actually matters to end user Presenter Synthesis Prompt You are the Presenter. Synthesize the Boardroom analysis into a clear, actionable final answer. ORIGINAL REQUEST: {USER_PROMPT} Boardroom ANALYSIS: [Step 1 - Strategist] Model A: {bullets} Model B: {bullets} Model C: {bullets} [Steps 2-5...] {all_council_responses} YOUR JOB: Write a clear, conversational response that: - Provides the best actionable answer - Incorporates key insights from all perspectives - Flags critical warnings/watch-outs - Asks 2-3 high-leverage follow-up questions (if needed) Tone: Professional but conversational. No meta-commentary about the Boardroom process. Recommendation Ship Phase 1 as MVP. The core value (multi-model synthesis) is immediately usable and differentiated. Notes visibility and advanced features can iterate based on user feedback. Timeline: 3-4 weeks to production-ready MVP. Resource ask: 1 backend engineer, 1 frontend engineer, light PM oversight. Expected impact: 15-25% of power users adopt within 60 days, measurable reduction in prompt iteration cycles. Questions for product/eng review: Do we gate this behind a plan tier or ship to all users? Do we want usage analytics per Boardroom execution? Should council models be configurable (admin settings) or hardcoded? Prepared by: Tom Schreier / TopAItoolsfor Date: 02/24/2026 Prepared using Poppy 😉

William Tom Schreier 9 days ago

💡

Feature Request