A fraud alert that reaches the handler before the transfer is authorised. A missed compliance step, prompted before the call ends. A customer about to cancel, flagged before they say the words. Conversation Intelligence turns live calls into structured, actionable data, at the carrier layer, in real time. And to be clear about who holds the switch: it applies only where you enable it - per trunk, per number, per call - and off means off.
Media is tapped passively at the network layer and mirrored into the AI pipeline. Nothing touches the call, nothing adds latency, nothing routes through third-party infrastructure. We can, because we're the carrier and this is our network.
An Operator is a natural-language prompt plus a declared output type. It watches the transcript as the conversation unfolds and emits typed signals: fraud.score, intent.detected, compliance.flag. Define your own for your own use cases.
A webhook delivers the signal to your platform mid-call, with the context to act on it. A whisper injects spoken guidance that only the intended ear hears - the handler's, or the caller's. The conversation never stops.
Every call becomes a vCon, an open IETF-standard record in your own S3-compatible storage. Memory is queried before the first word, so prior fraud signals, unresolved complaints and old context are already on the screen. We don't want your data.
Operators score fraud confidence turn by turn and fire a webhook the moment it crosses your threshold. We've delivered mid-call fraud signals to major banks for years; now the AI understands context, not just metadata.
fraud.score → webhookProtocol and disclosure monitoring as the words are said. A missed step triggers a whisper to the handler, not a finding in next quarter's audit.
compliance.flag → whisperSentiment, intent, compliance and resolution scored on every conversation, human and AI alike. A 2-5% QA sample becomes automated coverage of the lot, feeding CRM and dashboards without manual review.
sentiment.shift → dashboardA caller with documented history should trigger a different response than an unknown first-time caller. Memory surfaces it before the operator says hello.
memory.recalled → screen-popIt turns live calls into structured, actionable data at the carrier layer in real time. Operators emit typed signals, webhooks and whispers act mid-call, and every call becomes a vCon memory record.
No. Media is tapped passively at the network layer and mirrored into the AI pipeline - nothing touches the call, adds latency, or routes through third-party infrastructure, because Simwood is the carrier and it is our network.
Yes. You enable it per trunk, per number or per call - and off means off.
An Operator is a natural-language prompt plus a declared output type. It watches the transcript as the conversation unfolds and emits typed signals such as fraud.score, intent.detected or compliance.flag. You can define your own.
Every call becomes a vCon, an open IETF-standard record held in your own S3-compatible storage, and memory is queried before the first word. Simwood doesn't want your data.
Yes. Operators score fraud confidence turn by turn and fire a webhook the moment it crosses your threshold - the kind of mid-call fraud signal Simwood has delivered to major banks for years.
Live on the network now. The interactive documentation is the fastest way to see it working: real endpoints, real signals.