Answers from your documents, with a person ready when it can't
The assistant answers from your uploaded docs, website and FAQs, on your website widget and on WhatsApp. When it can't find a supported answer, it says so and hands the chat to your team.
Routing and context
Send each question to the right agent, with its own instructions and documents.
Agents per department
Set up separate agents for sales, support and billing. Each one has its own documents, instructions and tone of voice.
- Custom instructions per agent
- Keyword and intent routing
- Per-department knowledge bases
- Turn an agent on or off in one click
Conversation context
The assistant keeps the conversation history in view, so customers do not have to repeat themselves.
- Conversation history in every reply
- Customer context across sessions
- Source tracking on answers
Intent triage
Incoming questions are sorted by topic and sent to the right agent or to the human queue.
- Intent detection
- Quick path for greetings
- Topic categories
- Fallback routing
Answers from your documents
The assistant answers from what you upload, and says so when it cannot find an answer.
Your documents as the source
Upload PDFs and Word files, or add your website link. The assistant searches only your content when it answers.
- PDF, DOCX and web page indexing
- Semantic search over your content
- Per-agent or shared documents
Confidence thresholds
If the match with your documents is too weak, the assistant does not guess. It says it could not find an answer and offers a person.
- Answers only from your documents
- Says so when it cannot find an answer
- Source citations on answers
- Human handoff when unsure
Live chat and handoff
Replies stream as they are written, and your team can step in at any point.
Streaming replies
Answers stream word by word, so customers see the reply while it is being written.
- Word-by-word streaming
- Typing indicators
- Works on mobile and desktop
Human takeover
Your support team can watch live conversations, claim one, and take over from the AI.
- Claim a conversation in one click
- Live conversation viewer
- Hand the chat back to the AI
- Resolved conversations are logged
WhatsApp support
Run the same assistant on WhatsApp through the Meta Cloud API, including voice notes and reply buttons.
- Meta Cloud API webhook
- Voice note transcription
- Quick-reply buttons
- Human handoff on WhatsApp
Safety and privacy
Checks run on the server before a reply reaches your customer.
Server-side guardrails
Rules run before and after the model answers. They block prompt injection, filter off-topic requests and keep replies on-brand.
- Personal data masking
- Brand voice and safety rules
- Audit logs
Learning from your team
When a person on your team answers a question the assistant could not, that answer can be saved to the knowledge base for next time.
- Q&A saved from resolved chats
- Duplicate questions are skipped
- Shared across your agents
Encryption and privacy
Stored API keys, secrets and uploaded knowledge files are encrypted with AES-256-GCM. Your data is never used to train public AI models.
- AES-256-GCM for keys, secrets and files
- Role-based access control
- Not used to train public models
Operations
See how conversations go, pick your model, and send events to your own tools.
Analytics and CSAT
Track customer satisfaction ratings, resolution stats and first response times from one view.
- CSAT ratings
- Resolution stats
- First response time
Your choice of model
Bring your own API keys for OpenAI, Anthropic Claude, Google Gemini, DeepSeek, or any OpenAI-compatible endpoint.
- OpenAI-compatible providers, including Groq and Mistral
- Bring your own keys (BYOK)
- Automatic fallback to another provider
Webhooks and REST API
Send conversation events and escalations to your CRM, help desk, or any tool that accepts webhooks, as signed HMAC-SHA256 HTTP POSTs.
- HMAC-SHA256 signed webhooks
- Event subscriptions
- Works with any tool that accepts webhooks
How an answer is put together
A multi-step answer flow, document search in PostgreSQL, server-side safety rules, and your choice of model provider.
Every question goes through the same checks
Each question passes through several steps before a reply is sent: working out what the customer wants, finding the facts in your documents, checking the answer is supported, and running the safety rules.
- A fixed answer flow with fallback paths
- Separate agents for support, sales, billing and human escalation
- Replies stream to your customer word by word
Technical details
Answers come from your documents
The assistant searches the files, manuals and FAQs you upload and answers from them. When it cannot find a supported answer, it says so and offers a person.
- PDF, DOCX and website links are indexed for search
- A confidence threshold stops ungrounded guesses
- Source citations on supported answers
Technical details
Safety rules run on the server
Safety rules are applied on the server before answers are sent. Off-topic requests, prompt injection attempts and personal details are filtered out.
- Personal data masking
- Brand voice and tone rules
- Human handoff when the customer asks for a person
- Audit logs for review
Technical details
Send conversation events to your own tools
Conversation events and escalations are posted to your CRM, help desk, or any tool that accepts webhooks as they happen.
- Events: conversation created, message received, handoff, closed
- Signed webhooks, so your system can verify each delivery
- Works with any tool that accepts webhooks or a REST call
- Built-in test events and delivery stats
Technical details
Bring your own AI model and API key
Connect OpenAI, Anthropic Claude, Google Gemini, DeepSeek, Groq, or any OpenAI-compatible endpoint with your own API keys.
- Works with OpenAI, Claude, Gemini, DeepSeek and Groq
- Your API keys are stored with AES-256-GCM encryption
- Automatic fallback to another provider when one fails
Technical details
Put the assistant to work for your team
Paste your website link or upload your docs, then copy one embed snippet.