Machine intelligence
Models that run on the way in
Three instruments run on a message as it arrives, and the rest work on a schedule or when you open a thread. Deal terms come out, voice notes turn into text, and promotions get held back.
8
Instruments on ingest
42
Translation languages
350ms
Median processing time
0
Shared training pools
Core instruments
Eight instruments working inside the feed
Incoming messages pass through the chain in order. The instruments work out reply deadlines, catch promises made to a client and clear out noise, so your team reviews the conversations that matter.
Spam shield
Catches phishing links, crypto solicitations and bulk marketing drops before any notification fires.
Blocks 99.4% of automated cold spam at the gateway
Noise classifier
Separates two-factor codes, auto-replies and platform notifications from real conversations with buyers.
Routes verification codes to a throwaway sidecar tab
Thread summaries
Writes a four-bullet brief of a two-hundred-message group chat by the time you open the thread.
Rebuilt on every tenth inbound message
Commitment tracking
Pulls out promises like “I will send the revised quote by 4pm” and puts them on your calendar.
Flags an unfulfilled promise four hours before deadline
Reply debt
Counts the outstanding questions waiting on your answer across the networks you have connected.
Ranks unread contacts by historical trade volume
Relationship health
Watches for message frequency drops and tone shifts with long-term trading accounts.
Alerts the account owner when a weekly buyer goes quiet
Voice transcription
Turns an incoming voice note into plain text paragraphs the moment it lands in the thread.
Handles WhatsApp voice notes and regional audio formats
Media understanding
Reads text off photos of paper manifests, assay certificates, bank transfer slips and export documents.
Pulls invoice numbers and amounts straight into search
Writing assist
Draft replies in your own voice
Write faster while still sounding like yourself. The system reads your past outgoing messages and proposes replies that match your vocabulary and your punctuation.
Smart reply
Three context-aware reply options in one click, weighed against current inventory and open order status.
Inline autocomplete
Suggests the next four words as you type. Press Tab to accept the completion or keep typing to ignore it.
Voice reply
Speak your raw thoughts into the app on a walk. The engine cleans the audio into a formatted business note.
Live translation
Talk to international suppliers in their language while reading and typing in your own. Forty-two languages.
Live translation, 42 languages
The customer writes in Arabic, you read it in English, you type your answer in English, and they receive Arabic. Both the original text and the translated version stay in the thread.
received • ar
هل وصلت الشهادة؟
“Did the certificate arrive?”
sent • en → ar
“Yes, emailed this morning.”
PersonaLearn studies your past replies to write a full message in your voice. It has its own page, and its own privacy rules.
How PersonaLearn worksMedia understanding
Read documents inside incoming images
Clients snap photos of diamond grading reports, customs declarations and bank receipts rather than sending a clean PDF. The media engine pulls the serial numbers, the amounts and the stamps out of the picture so you can search them later.
Assay certificates
Extracts carat weights, cut grades and lab numbers from phone camera snaps.
Wire receipts
Pulls sender name, IBAN and total sum from mobile banking screenshots.
Shipping manifests
Reads stamped paper airway bills and matches tracking codes to open deals.
Voice memos in dialect
Transcribes hurried voice notes recorded in noisy trade halls and ports.
Safe deployment
Test the AI in shadow mode
A model should not reply to a real customer until you have checked its work. Shadow mode replays your last five hundred closed conversations through the engine in a sandbox. You read each proposed draft next to what your team sent, then decide.
Pick the history
Select five hundred closed threads from WhatsApp, Telegram, Signal or email.
Run the simulation
The engine processes every inbound turn and drafts a response in the background.
Read the report
See where the model matched your team, where it held back and where it was wrong.
Turn on live assist
Enable suggested drafts for your staff with the confidence thresholds you verified.
Replay, 500 conversations
simulationFive hundred threads replayed. 418 matched what your team sent, 74 were held for a human, 8 you would have rewritten.
98.1% of the 426 send decisions needed no edit
Data boundaries
Your communication data stays private
Messages are processed inside your own tenant. Trading terms, contact details and price negotiations never enter a public training set or a shared model memory.
Base model training excluded
Your messages are never used to train a base model for us, for Anthropic, or for anyone else.
No multi-tenant data pooling
Data from your workspace is separated at the database level and never combined with another company's records.
No human message review
Our engineers cannot read your inbox contents or open your attachments while working a support ticket.
Isolated custom keys
Bring your own provider keys so inference runs under your own enterprise agreements.
Ephemeral prompt handling
Inference requests are discarded once processed. Nothing lingers on an intermediary cache server.
Where it runs
Models execute in dedicated European instances. Encryption keys stay in your custody through bring-your-own-keys.
Security detailMore of the platform
Unified inbox
Nineteen networks in one shared team timeline.
Total recall
Permanent retention and search across your connected channels.
PersonaLearn
Drafts written privately in your own voice.
Export and migration
Exports up to your plan ceiling and one-pass migration from 12 systems.
Apps
Native clients for macOS, Windows, iOS and Android.
Integrations
Sync for CRMs, billing systems and ticketing.
Run triage on every channel
Start with a five-hundred-conversation shadow run. Read the output before your team sends a single live message.