Documentation
Your conversational AI agents — understand, respond, and take action on WhatsApp.
A bot is your conversational AI agent. It reads incoming WhatsApp messages, reasons about them using an LLM (large language model), and replies — drawing on a knowledge base you provide, tools you enable, and a system prompt that defines its personality and boundaries. A bot is not the phone number itself; it is the brain behind the number. One bot can power multiple WhatsApp numbers, and switching the number behind a bot does not break existing conversations.
Conversations are tied to the bot + the contact, not to a specific WhatsApp number. When a message arrives on any number assigned to the bot, Neuron looks up the contact and resumes the existing conversation — with full message history intact. This means: • If you disconnect one number and connect a different one, conversations carry over seamlessly. The contact keeps their full history. • If a bot has multiple numbers, a contact always sees one continuous conversation regardless of which number delivered each message. • A human agent can take over a conversation (takeover), send replies from the dashboard, and hand it back to the bot — all within the same thread. The conversation is the durable thread; the number is just the delivery channel.
Navigate to Bots and click "Create Bot". You'll need to provide a name and a system prompt. The system prompt is the most important configuration — it defines your bot's personality, tone, scope, and behavior. You can also use the AI Setup wizard to describe what you want in plain language and let Neuron generate the system prompt, knowledge base entries, and tool configuration for you.
For a faster start, click "AI Setup" on the Bots page. Describe what you want your bot to do and Neuron generates the full configuration — system prompt, knowledge base, and tool setup.
The system prompt is injected at the start of every AI conversation. Write it as instructions to the AI. Be specific about what the bot should and shouldn't do, its tone of voice, and how it should handle edge cases. The prompt has access to context variables — the contact's name, phone, conversation history, and any knowledge base results — so you can make it dynamic.
Include examples of ideal responses in your system prompt. The AI performs better when it has concrete examples to follow.
Choose from available LLM models on OpenRouter. Different models have different strengths — some are faster and cheaper, others are more capable at complex reasoning. The temperature setting (0.0–2.0) controls creativity: lower values give more consistent responses, higher values give more varied ones. You can also bring your own API key (BYOK) for a specific provider if you prefer.
Configure three special messages: the Greeting Message is sent automatically when a new conversation starts (a new contact messages the bot for the first time), the Fallback Message is sent when the AI cannot generate a response, and the Escalation Prompt defines when and how the bot should hand off to a human agent.
Two toggles control where a bot replies: process direct messages and process group messages — both are off by default, so turn on what you need. For groups, you can further restrict the bot to a whitelist of specific groups. Incoming messages are always saved to the conversation history either way; these toggles only control whether the AI responds.
A bot can be assigned to more than one WhatsApp number (session). This is useful for: • **Higher throughput** — spread outbound messages across several numbers to avoid WhatsApp rate limits. • **Geographic or brand separation** — different numbers for different regions or product lines, all handled by the same AI brain. • **Resilience** — if one number goes offline, the others keep working. When multiple numbers are attached, Neuron load-balances outbound sends across them (round-robin, least-used, or random) with sticky per-contact affinity — each contact keeps talking to the same number so the conversation feels natural. Manage secondary numbers and the balancing strategy from the bot's detail page.
Conversations are bot-scoped, not number-scoped. You can freely add, remove, or swap numbers without losing any conversation history. The bot and the contact's thread persist regardless of which number is attached.
Beyond chatting, a bot can take actions with tools. Neuron ships dozens of built-in tools — send images and documents, search the knowledge base, look up and save contacts, create polls, set reminders, escalate to a human, post WhatsApp Status, and more — which you switch on or off per bot in the "Built-in Tools" tab. You can also add your own custom tools (HTTP APIs, webhooks, or MCP servers) in the "Tools" tab. See the Tools section for the full catalog and setup.
A bot can run autonomous tasks on a schedule — a one-off follow-up, a recurring check-in, or an interval-based nudge. Create and manage them in the bot's "Tasks" tab: give the task a title, an instruction (what the bot should do), and a schedule (once, recurring by cron, or every N minutes/hours). Each run executes the instruction with the bot's tools and is logged with its output. The bot can also schedule tasks for itself mid-conversation when you enable the "Create Scheduled Task" built-in tool.
Each bot can have API keys for programmatic access. Generate them from the bot's API tab — they start with "nrn_" and let external applications send messages, get AI responses, and list conversations via the Bot API. The Playground tab lets you test conversations with your bot directly in the browser without needing a WhatsApp number connected.
You can pause a bot at any time. A paused bot will not respond to new messages — incoming messages are saved to the conversation history but the AI does not process or reply to them until the bot is resumed. This is useful for maintenance or when you want to temporarily take full manual control.
Silent mode is a per-session setting. When a session is silent, the bot still runs the full pipeline (LLM and tools) but does not send a visible reply — useful for group moderation, where the bot watches and runs tools without posting chat messages. The assistant's output is still saved and visible in the dashboard.
Bots can learn from your conversations. When configured, the bot analyzes conversations and proposes improvements — reflections — that you review and approve before they take effect. Approved reflections are synced into the bot's knowledge base, so it gets smarter over time without you rewriting the system prompt. Configure learning sources and review reflections from the bot's Reflections tab.
A persona captures a distinct writing voice — tone, diction, emoji habits, themes. Build personas under the Personas section (you can distill one from a collection of sample content) and apply one to a bot so its replies keep a consistent voice. Personas are managed separately from the bot so the same voice can be reused across multiple bots.
Each bot has an Analytics tab showing message volume, conversation counts, response times, and escalation rates. Use these metrics to monitor performance and identify areas for improvement.