Conductor overview

Conductor is an assistant inside the dashboard, and it is the first thing you see: Home in the sidebar is Conductor. You describe what you want in plain language and it acts on your workspace:

• Build and edit agents. Build a new agent from a short conversation, or propose changes to an existing one as diffs you accept or reject.
• Test. Write simulation test cases for an agent and run them as a batch.
• Investigate calls. Read selected or recent calls and summarise what went wrong.
• Build dashboards. Create an analytics dashboard from a request such as "call volume and latency by agent".
• Guide you. Produce step-by-step walkthroughs, and run them as guided tours that highlight each control on the page.

Needs keys Conductor uses OpenAI. Without a key it runs in mock mode: it picks actions with keyword rules and uses built-in templates, so every feature works for a demo but the content is generic.

Where to find Conductor#

Home#

Home greets you by name and asks How can I help?. Type anything into the composer, or start from a quick start:

Conductor History on the left lists your saved conversations, newest first, with Search chats and New chat. The old /conductor address opens Home.

Two kinds of Conductor conversation#

Conductor has two back ends. They look alike but do different things.

Actions in a Conductor conversation#

Every message runs one action. With auto (the default), Conductor picks it from your message.

Attach context#

Conversations can carry up to 10 attachments in total:

In the side panel, the paperclip opens Attach context with search tabs for Calls, Dashboards, Cohorts, Contacts, Agents and Nodes. On Home, the + button in the composer opens the same picker, and the paperclip next to it attaches a text file.

Attachments stay on the conversation, so later messages can refer to them.

Guided tours#

The side panel has a Guided tour box. Type a topic ("set up AI QA") and choose Show me, or choose Start guided tour under a walkthrough reply. The tour moves to each step's page and highlights the control, with Back, Next and Done.

Built-in tours cover connecting a phone number, adding a knowledge base, running a batch call, setting up AI QA, creating an alert, receiving webhooks, reviewing calls, building an analytics dashboard and setting up the workspace. Any other topic gets the "Build and launch an agent" tour. With an OpenAI key, Conductor can also write steps for other topics.

Credits#

Each message in a Conductor conversation costs credits by model tier:

Choose the tier in the composer on Home. The side panel always uses Default.

• Free: 30 credits per user per day and 200 credits per workspace per day. Days reset at midnight UTC. A message is free only if both allowances cover it.
• Pay-as-you-go: after the free credits, $0.30 per credit, up to a monthly limit. Turn it on in Settings, Reliability: Conductor pay-as-you-go and Conductor monthly limit (AUD). It is off by default.
• Out of credits: the API returns 402 with error_code conductor_credits_exhausted (free credits used, pay-as-you-go off) or conductor_limit_reached (monthly limit reached).

Home shows your free credits left today next to the composer. Get Conductor credits returns the full picture: yours and the workspace's free credits left, and pay-as-you-go spend this month. Each message is charged as free_user, free_workspace or paid, and paid credits appear on the workspace's usage.

API#

curl -X POST https://api.tried.com.au/conductor/sessions/$SESSION_ID/messages \
  -H "Authorization: Bearer $TRIED_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "content": "Why did my recent calls fail?",
    "tier": "default",
    "attachments": [{ "type": "call", "id": "call_b734725b6c060c5764b31803" }]
  }'

{
  "reply": "I looked at 1 call. 0 had no problems. The most common issues: not successful (1), negative sentiment (1).",
  "action": "investigate_calls",
  "result": { "type": "investigation", "calls": [], "patterns": [{ "issue": "not successful", "calls": 1 }] },
  "mode": "llm",
  "credits": { "charged": 1, "source": "free_user", "cost": 0 },
  "credit_status": { "tiers": { "fast": 0.5, "default": 1, "advanced": 3 } }
}

mode is llm when OpenAI planned the turn and mock when the rules did (including when the OpenAI call fails). The examples are trimmed.

Next#