Skip to Content
AIOverview

AI in HeyKazi

HeyKazi’s AI is built around one idea: AI grounded in your firm’s own data, under your control. It reads your matters, clients, trust ledger, time, and compliance state to produce work that reflects how your practice actually operates — and it never acts on anything sensitive without an attorney’s approval.

Three principles run through every AI feature:

  • Bring Your Own API Key (BYOAK). AI runs on your organisation’s own Anthropic key, so you control cost, usage, and data egress. See Integrations to set it up.
  • Draft-first, human-approved. AI proposes; a person decides. Consequential actions are held in an approval queue for attorney sign-off before anything changes.
  • Grounded and SA-tuned. Output is shaped by your AI firm profile (practice areas, house style, risk posture) and South African legal knowledge (FICA, the Legal Practice Act, LSSA tariffs).

Two ways to use AI

1. Inside HeyKazi (AI on your data)

  • AI Assistant — a conversational helper that answers questions, looks up your data, and drafts content from any page.
  • AI Specialists — focused, one-shot skills for legal tasks: FICA verification, matter intake, contract review, drafting, and compliance audit. Each returns structured, cited output and routes consequential actions through the approval queue.

Both run server-side on your Anthropic key and are metered against your monthly budget.

2. Bring Your Own Claude (your Claude, Kazi’s context)

If your team already uses Claude (Claude Code, Claude Desktop, or claude.ai), you can connect it directly to your HeyKazi data over a secure, read-only connection — see Bring Your Own Claude. Your firm’s own Claude becomes grounded in your live matters and trust ledger, and your own Claude subscription pays the token bill.

The two paths complement each other: in-product AI is HeyKazi calling Claude (and writing back, gated); Bring Your Own Claude is your Claude reading HeyKazi (never writing). Both are scoped to what each user is allowed to see, and both are fully audited.

Safety, privacy, and audit

  • Scoped to the user. Every AI feature only ever reads data the current member is permitted to see — the same role and capability checks as the rest of the app.
  • Audited. AI activity is recorded in the audit log (the ai.specialist.* events for in-product AI; read events for Bring Your Own Claude), giving you a defensible trail of what AI touched.
  • POPIA-aware. Sending client data to an AI provider is a deliberate, opt-in choice — controlled by your API key for in-product AI, and by an explicit data-egress consent for Bring Your Own Claude.
AI AssistantAI Specialists