Cynosure is a multi-agent AI orchestration system — a Palantir-style command centre that routes your inputs to specialized AI agents and surfaces structured intelligence back to you via Telegram, Discord, or this dashboard.

You send a message. Cynosure decides which agent handles it, what tools to use, and how many agents to chain together. Every decision is logged, scored, and fed back into the routing engine so it gets smarter over time.

Agent Selector

Auto-detect — routing engine picks the best agent from your message keywords.
Force a specific agent by selecting it from the dropdown, or mention them explicitly in your message (@atlas, @edge, etc.).

Mode Selector

Mode tags are prepended to your message. HERETIC routes to an abliterated model (Dolphin Venice) — for queries that standard models refuse or over-hedge on.

Inline Indicators

Every response is automatically scanned and annotated:

• Percentages — progress bar (green >5%, amber 3–5%, red <3%)
• Confidence — ● High / ● Medium / ● Low
• Odds — +150 -110 cyan pill

Visual Analysis

Include a trigger phrase in your message to auto-generate a Chart.js panel below the response:

visualize this · chart this · show me a chart · graph it · visual analysis · break it down visually

Or click ▸ Expand Analysis on any response where numerical data was detected.

Agent Mentions

Mode Tags (prepend to any message)

Meeting Commands (Atlas)

Auto-routing (no mention needed)

Just send a natural language message. The routing engine reads keywords and routes automatically. "Full analysis on Lakers tonight" → Edge. "Review my HVAC submittal" → Atlas. "Write a TikTok hook" → Scribe.

How It Works

Paste a raw transcript (voice memo, Otter.ai export, rough notes) into the Meetings tab. Atlas processes it via OpenRouter (claude-haiku-4-5) and returns structured output in seconds.

What Atlas Extracts

• Cleaned minutes — structured into Attendees, Agenda, Discussion, Decisions Made
• Action items — each with owner, due date, and priority (HIGH/MEDIUM/LOW)
• RFI triggers — any item needing formal clarification flagged with responsible party and spec section
• Email drafts — follow-up emails for decisions requiring external communication

Flow to Kanban

After processing, click + Add All to Kanban to create a todo for every action item — automatically assigned to the right agent based on owner name.

Meetly Webhook (future)

When Meetly goes live, it will POST to POST /api/webhook/meetly on the bridge. The endpoint is wired and ready — just needs the Meetly payload schema mapped in bridge.js.

How Edge Logs Picks

Every analysis Edge produces is automatically logged to the betting_log Supabase table. The Betting tab reads this table and displays your full record.

What the Stats Mean

Edge's Analysis Structure

Every analysis follows 6 layers: Vegas lines → Team form/ATS → Injuries/props → H2H trends → Edge % → CLV/sharp money. Edge always uses web_search to pull current odds before responding.

Trigger Phrases

Include any of these in your Command tab message to auto-generate a chart below the response:

visualize this
chart this
show me a chart
graph it
visual analysis
break it down visually

Chart Types (auto-selected)

Expand Analysis Button

Even without a trigger phrase, if the response contains structured numerical data (e.g. a breakdown of stats, percentages, comparisons), a ▸ Expand Analysis button appears. Click to render the chart on demand.

Example Prompts That Work Well

@edge Full NBA slate analysis tonight. Visualize this.
@vault Compare QQQ, SPY, VTI year-to-date. Chart this.
@atlas Budget breakdown by trade. Break it down visually.