AI-Powered Collaboration

AI That Respects Boundaries

Artificial intelligence is transforming how teams work. Automated transcription turns meetings into searchable records. Summarisation distils hour-long calls into concise action points. Smart replies reduce the friction of routine messaging. These capabilities are genuinely useful.

They are also, in many implementations, genuinely problematic. When your meeting audio is processed by a US-hosted AI service, the content of your confidential discussions has left your jurisdiction. When transcripts are stored indefinitely to train a provider’s models, your intellectual property becomes someone else’s training data. When AI features are enabled by default without participant consent, privacy expectations are violated silently.

Mandraki takes a different approach. AI features are powerful, transparent, and governed by the organisation — not imposed by the platform.

What Mandraki’s AI Can Do

Meeting Transcription

Mandraki can transcribe video calls in real time, producing a text record of the discussion that is synchronised with the call timeline. Transcripts are searchable, exportable, and stored with the same encryption protections as all other organisation data.

Transcription supports multiple languages and can distinguish between speakers, making it straightforward to review who said what during a complex discussion.

Call Summarisation

After a meeting concludes, Mandraki can generate a structured summary that captures the key discussion points, decisions made, and action items identified. Summaries are concise and designed to be shared with colleagues who could not attend, reducing the need for lengthy follow-up emails.

Smart Replies

In messaging, Mandraki can suggest contextually appropriate replies based on the conversation history. Suggestions are generated on-device where possible and via EU-hosted AI processing where additional context is required. Smart replies are suggestions only — they are never sent automatically.

Recording Analysis

For organisations that record meetings, Mandraki’s AI can analyse recordings to extract topics discussed, sentiment patterns, and participation metrics. This is particularly valuable for training, compliance review, and organisational learning.

Where AI Processing Happens

All AI processing runs on European hyperscale infrastructure. Audio, text, and metadata are processed within the EU and are never transmitted to non-European servers. There is no dependency on US-based AI APIs, and no data is used to train third-party models.

This is not a contractual promise backed by Standard Contractual Clauses. It is an architectural fact. The AI compute infrastructure is European-owned, EU-located, and operated under European law.

Governance and Consent

Mandraki provides three levels of AI policy that organisation administrators can configure.

Disabled. AI features are turned off entirely. No audio is transcribed, no messages are analysed, and no summaries are generated. This mode is fully compatible with end-to-end encryption.

Opt-in. AI features are available but require explicit consent from all call participants before activation. When a participant initiates AI transcription, every other participant receives a clear notification and must consent before processing begins. Consent is logged with timestamps and feature-level granularity.

Enabled. AI features are active by default for all calls and channels. Participants are informed that AI is active, but individual opt-out is not required. This mode is appropriate for organisations that have obtained blanket consent through their employment or service agreements.

AI and End-to-End Encryption: A Deliberate Choice

End-to-end encryption and server-side AI processing are mutually exclusive by design. If a call or channel has end-to-end encryption enabled, AI features cannot be activated, and vice versa. This is not a technical limitation to be worked around — it is an honest acknowledgement that you cannot have a server process content that the server cannot read.

Organisations can make this choice on a per-call or per-channel basis. A board meeting might use E2EE with no AI. A project standup might use AI transcription without E2EE. The choice belongs to the organisation and is enforced technically, not merely by policy.

Data Retention

Organisations control how long AI-generated content is retained. Three retention modes are available: transient (deleted immediately after delivery), session (retained for the duration of the call), and persistent (retained according to the organisation’s data retention policy). Organisations using Bring Your Own Key encryption are automatically restricted to transient retention, ensuring that AI outputs cannot outlive the customer’s control of their encryption keys.