Internal Notes

Internal notes — labeled Notes in the admin panel — are a structured memory space scoped to a conversation, fully invisible to the end user. They let the agent remember facts it picked up during the chat and give the human team a quick context view without having to re-read the whole transcript.

What they are

Every conversation carries a notes object in key-value format. Keys are defined by the agent itself (or by an admin during intervention), and values can be text, numbers, booleans, or structured JSON objects.

Common examples:

• customer_name: "Maria Silva"
• product_in_question: "Pro annual plan"
• mood: "frustrated"
• refund_attempts: 2
• next_step: "schedule callback on Friday"

Notes are visible in the admin panel, in the Notes card below Info (to the right of the chat). The end user never sees them. When the conversation has none, the panel shows “No notes yet”.

Who sees what

Persona	Can see notes?
End user (Hub, WhatsApp, public chat, widget, etc.)	No
Organization admins	Yes — side panel next to the chat
The agent itself (LLM)	Yes, if the Read Notes tool is enabled

In other words: internal notes are a side channel between the agent and the human team. The user keeps having a normal conversation; in parallel, the agent writes down what it considers important and the team can check anytime.

How the agent reads and writes notes

There are two dedicated native tools:

Read Notes (read_notes)

Lets the agent read all notes already saved on the current conversation. No configuration — just enable.

Use it when the agent needs to:

• Recall information collected in earlier messages.
• Avoid re-asking for data the user already provided.
• Adapt tone based on an earlier signal (e.g. mood: frustrated).

Write Notes (write_notes)

Lets the agent create or update keys on the notes object. Also configuration-free.

Use it when the agent needs to:

• Record structured data that came up in the conversation (name, phone, product).
• Track progress through a multi-step flow.
• Leave context for a human who may take over later.
• Persist the outcome of a decision so it doesn’t redo work on the next turn.

Enabling the tools on your agent

1. Open Agents -> [agent] -> edit -> Tools.
2. Add Read Notes and/or Write Notes.
3. In the agent’s instructions, tell it when to use each one. Example:

“Whenever the user provides a name, phone, or problem, save it to notes with write_notes. At the start of each reply, consult notes with read_notes.”

Without prompt guidance, the model tends to use notes sporadically. Be explicit.

Use cases

• Long-running support: a conversation that spans days. The agent records what’s resolved and what’s pending; when the user returns, it picks up where it left off.
• Multi-channel support: the user starts on WhatsApp and moves to web chat; notes travel with the conversation.
• Human handoff: before calling the Human Intervention tool, the agent writes a summary to notes (escalation_reason, attempts_made) — the admin taking over has instant context without re-reading the full transcript.
• Lead qualification: the agent fills in lead fields (interest, budget, buying moment) that the sales team later consumes.

Good practices

• Use short, consistent keys across conversations (customer_name, not nameOfTheCustomerWhoJustArrived). This helps when you want to analyze multiple conversations.
• Don’t duplicate in the prompt what’s already in notes. The agent reads on demand — it saves tokens and keeps context fresh.
• Don’t store raw secrets in notes. Notes aren’t encrypted and are visible to any admin in the organization. For sensitive data (passwords, tokens), use the secure credential integrations.
• Review notes in model conversations before tweaking the prompt — they show what the agent is actually capturing from reality.