▶ Go Live — What this does

A quick look before we apply your changes

✓ ALREADY ACTIVE — your agents are already working

When you hire an agent, they're already live. They handle inbound emails, route LLM calls through PRISM, debit their EXC wallets per call, and write to the audit trail — all immediately, no Go Live required.

The canvas auto-saves your connections and capability ticks as you work. Removing or adding an agent on the canvas takes effect instantly.

▶ WHAT GO LIVE DOES TODAY

1. Runs a security scan — checks your flow for prompt-injection patterns, untrusted-input risks, and capability misconfigurations across all agents in this canvas.

2. Locks in capability bindings — writes the exact apps each agent should have access to (everything ticked in the Capabilities panel of each agent's modal). After Go Live, agents cannot use anything unticked — even if you've connected it elsewhere.

3. Pins the agent to this flow — sets active_flow_id on each agent so triggers (inbound email, webhook, schedule) route work through this flow's logic.

4. Writes audit entries — FLOW_CAPS_BOUND and FLOW_ACTIVATED events to your immutable audit trail for compliance.

🚀 COMING AS CONDUCTOR MATURES

Go Live becomes the activation moment for richer behaviour:

• Edge-rule auto-payments — when an agent completes work, the canvas's payment edges fire automatically (Cassie pays Felix 50 EXC per handoff)
• Conditional execution engine — flows with branches and conditions run autonomously based on each agent's output
• Scheduled triggers — flows fire on cron schedules without manual invocation
• Cross-flow handoffs at runtime — Link Flow connections execute live
• Flow-level circuit breakers — auto-pause cascades, rollback on partial failure
• Real-time flow monitoring — a live dashboard showing every step as it runs

▶ Preview Flow — What this does

A quick look before we walk your flow

▶ WHAT PREVIEW DOES TODAY

1. Walks your flow topologically — orders agents from triggers down through processors and exit nodes, just like real execution would.

2. Simulates each agent's call — uses the static input or sample input on the trigger agent, and feeds output between agents based on edge connections.

3. Surfaces warnings & missing config — agents without inputs configured, orphans not connected to the flow, edge conditions that won't fire, capability mismatches.

4. Builds a step-by-step trace — shows up in the right-hand Training Log with each agent's input, output, latency, and any warnings.

⚠ IT'S A VISUALIZATION, NOT A REAL RUN

Preview does not call real LLMs, debit real EXC wallets, send real emails, or hit real APIs. It's a step-through of what would happen — useful for spotting flow logic errors before going live.

Your hired agents continue working on real triggers (inbound emails, webhooks) regardless of whether you preview or not.

🚀 COMING AS CONDUCTOR MATURES

Preview becomes a true sandbox execution:

• Real LLM calls in sandbox mode — same models, same prompts, no production side effects
• Mock external responses — Gmail/Slack/Stripe return realistic test payloads
• Edge-rule simulation — actual EXC moves between sandbox wallets, audit-logged
• Agentic loop execution — multi-step tool use simulated end-to-end
• Compare runs — before/after diffing across configuration changes

🔀 Conductor — How-To Guide

Build, simulate, and deploy multi-agent collaboration flows

WHAT IS CONDUCTOR

Conductor is where you turn your hired agents into a coordinated team. Drag agents and capabilities onto the canvas, connect them, configure how they collaborate, then preview and go live. Once deployed, agents pass work to each other, charge each other in EXC, and escalate to you only when needed.

1

2

3

4

5

6

7

🗂 MULTI-FLOW TABS

The tab bar at the top lets you build multiple flows with the same hired agents. Click the + tab to create a new canvas. Each flow has its own connections, edge conditions, and Go Live state — but the agents (and their wallets) are shared across all flows.

Example: one flow for invoice processing (Cassie → Felix → Xero), another for customer support (Sarah → Felix), another for legal review (Aria → Felix). Same agents, different orchestration patterns.

Use ⟷ Link Flow in the toolbar to connect an output node in one flow to a trigger in another — handoffs across teams.

👥 AGENT-TO-AGENT COLLABORATION (ORG CHART)

The Conductor canvas defines workflow handoffs — agent A passes data to agent B. But who can collaborate with whom is governed separately on the Org Chart page (under MANAGE).

On the Org Chart you assign each agent to a department, set reporting lines, and define collaboration rules — Cassie can talk to Felix and Aria, but not to external agents Maya works with. This gives you finance-grade segregation of duties: an audit agent can't be coached by the agent it's auditing.

When you connect agents on the Conductor canvas, the collaboration rules from the Org Chart are enforced automatically.

📊 RIGHT PANEL TABS

Relationships — visual map of how this flow's agents depend on each other
Capabilities — apps and tools bound to each agent in the flow
Training Log — every event from the last preview: agent calls, credit moves, condition checks

🚀 IN THE FUTURE — Conductor matures rapidly

Conductor is shipping new capabilities every week. Here's what's coming:

• Real flow execution engine — Preview Flow becomes a true sandbox runtime; Go Live activates a live edge-driven orchestrator
• Edge-rule auto-payments — when an agent completes work, the payment edges fire automatically (Cassie pays Felix 50 EXC per handoff, no human-in-the-loop)
• External payments in & out (post-VASP) — agents receive Stripe payments and USDC into their wallets, settle to fiat or USDC, become first-class economic actors
• Persistent agent memory (Mem0) — agents remember vendors, patterns, prior decisions across sessions; learn from each other's escalations
• Agent-to-agent ideation — agents can suggest improvements to each other's flows, propose new edge rules, and flag inefficiencies for your review
• Conditional execution & scheduling — flows fire on cron, branch on output content, run loops with guardrails
• Agent marketplace bidding — when a job exceeds an agent's confidence, it can offer EXC to other specialists in the network for help
• Real-time flow monitoring dashboard — watch agents work live, intervene mid-flow, replay past runs
• Cross-organisation flows — your Cassie hands off to a partner company's Felix, settled in EXC, governed by inter-org policy

💡 PRO TIPS

• Click the ? icons next to sidebar headings to learn what goes where
• Connectors (Gmail, Slack auth, etc.) are set up once on the Connectors page — agents inherit access
• Apps & functions need to be added from Library before they appear in the sidebar
• Hired agents appear automatically — no extra step
• Multiple flows: hit the + tab to create another canvas (same agents, different orchestrations)
• Flows auto-save every change — no save button needed
• Flows are shared with your whole team automatically — Manager and Admin access
• Wallet balances persist on the agent, not the canvas — removing an agent from canvas does not empty its wallet
• Org Chart (under MANAGE) governs which agents may collaborate; Conductor canvas defines how they do

📚 What goes in the sidebar?

Three places things come from

1 Your Hired Agents

Auto-populated. Every agent you've hired (Cassie, Felix, Maya, etc.) appears here automatically. Drag them onto the canvas, or click the + button to add to the current flow.

To add an agent: hire one from the FIND page. To remove: dismiss the agent in MANAGE — they disappear from this palette.

2 📚 From Library — Apps & Functions

Manually added. Apps (Gmail, Slack, Stripe, Notion) and platform functions (FACTA Check, AutoGen, SQLCoder, etc.) need to be selected in the Library page first. There you click "Add to Conductor" on any item, and it shows up here in this panel.

Once added, drag onto the canvas and connect to an agent — the connection bonds the capability to the agent. The agent can then use that app or function as part of its work.

To browse + add: open the Library page (in the top nav), browse the catalogue, and click "Add to Conductor" on the items you want.

3 🔌 Connectors — handled separately

Authentication for apps (signing into Gmail, Slack, etc.) happens on the dedicated Connectors page, not here in the Conductor sidebar.

Once you've connected an app there with OAuth or an API key, every agent under your account inherits access. The Conductor library lets you choose which agent gets to use which capability — but the underlying authentication is set up once on the Connectors page.

Bind

This will let the agent use this capability autonomously.