AI built for the way your firm actually works

Organize conversations, files, and team members into projects scoped to a client engagement or workstream.

Context persists across every conversation within the project. Assign standing instructions as system context for the whole project. Classification is enforced at the project level — a project floor can rise above its parent hierarchy level, but never fall below it. Project budget is tracked independently from the org layer: a blocked project does not block other projects or the org.

The chat console works the way your team already expects an AI tool to work — streaming responses, conversation history, file-ready design.

The difference is invisible until it matters: every message carries its classification tier, budget position is visible in-line, and everything is recorded in the audit trail. Nobody retrains; compliance still gets evidence.

Multiple people in the same AI conversation at the same time — real shared state, not screen-sharing a private chat.

An engagement team works a problem together: the partner steers, the analyst drills in, and the AI sees the whole thread. Presence and shared context are managed by a real-time collaboration service using conflict-free replicated state. Participant roles — Owner, Collaborator, Observer, Supervisor — enforce who can prompt and who can only watch, all through the same governance pipeline.

Fork a conversation at any message — like a git branch for your thinking — while the original thread stays intact.

Explore an aggressive tax position on a branch while the main line stays conservative. Compare two framings side by side. Every branch runs through the same governance pipeline as the main thread. Classification can escalate from the parent but never de-escalate.

Switch between models — Claude, GPT, and others your admin enables — inside one conversation, keeping the context.

Draft with one model, critique with another, without re-pasting context into a second tool. Every switch is logged: conversation ID, previous model, new model, token count at switch. Classification rules carry across: PHI conversations cannot switch to non-BAA models. If Conversation Optimisation Mode is on, the platform routes automatically — you can always override.

Share a conversation — or just a branch of it — with colleagues, clients, or auditors, with classification enforced at the moment of sharing.

Share recipients can read and comment; they cannot trigger inference or incur cost. The classification tier is snapshotted at share time and cannot drift. Revoke access at any moment — the share link stops working instantly. Every view, comment, and revoke is in the audit trail. GDPR export and erasure cover share records.

Surface the right context automatically — from your personal memory, your team's knowledge base, or both — without pasting it in every time.

User-level memories attach facts from past conversations that should travel with you. Org-level knowledge bases hold firm documents, policies, and precedents, searchable via vector retrieval (pgvector HNSW). The Knowledge Down principle: a KB at a higher hierarchy node is visible to all nodes beneath it. PHI memory and KB chunks only surface in PHI-classified conversations.

Long conversations hit context limits. Continuation Mode handles it automatically — summarising and picking up where you left off, audit trail intact.

Three settings: Off (manage length manually), Summarise & Start New (you review and edit the summary before continuing), or Auto-Continue (platform generates the summary and continues, with a 5-minute undo window). The context usage bar in the conversation header shows utilisation in real time. Continuation chains are atomic for GDPR erasure.