Now in beta

Gate Kali Assistant

A self-hosted, multi-tenant AI principal assistant. With your OAuth consent it reads Gmail, Google Calendar and Google Drive to produce briefings, meeting prep, triage and drafts — adapting tone and priorities through context lenses while every consequential action waits for your approval. Google access is limited to the scopes you approve and can be revoked at any time.

Try the beta All products

How it works

Three steps from signup to a governed assistant that knows your world.

Create your workspace. Sign up at mygate.gatekali.com — public signup provisions an isolated tenant with an admin account and white-label branding controls.

Connect your world. Link Gmail or IMAP/SMTP, Google or Apple calendars, Drive, documents, and messaging channels — WhatsApp, Telegram, Slack, Teams or Discord — via OAuth or link codes.

Delegate with control. The assistant briefs, triages, prepares and drafts. Reads run freely; sends, deletes and shares pause for your explicit approve/reject — every decision is audit-logged.

Capabilities

Built on Clean Architecture: a model-agnostic router (OpenAI, Anthropic), a Postgres + pgvector knowledge base, and a policy engine that keeps a human in charge.

Assistant intelligence

●

Context lenses

Seven built-in lenses — Personal, Diplomatic, Government, Security, Media, Commercial and Social Protocol — auto-detected from your message, channel and time of day to shape tone, priorities and tool access.

●

Executive briefings

Schedule, priority correspondence, deadlines, documents and business alerts synthesized into a morning briefing — or a meeting brief pulled from related emails and files.

●

Meeting intelligence

Bots auto-join meetings, record audio, transcribe and summarize. Prep pulls the relevant emails, documents and previous context before you walk in.

●

Memory, notes & reminders

Durable conversations with memory compression, tagged notes, and scheduled reminders swept by an atomic scheduler and pushed live over WebSocket.

●

Skills & proactive checks

Record, replay and share multi-step workflows as skills — and configure heartbeat checks for inbox, calendar and follow-ups.

●

Document knowledge base

Upload files; they're chunked, embedded and searchable through pgvector — source documents encrypted at rest.

Channels & integrations

●

Every channel you use

Real-time web chat plus WhatsApp, Telegram, Slack, Teams and Discord webhooks — with link-code pairing to connect a channel to your account.

●

Email, calendar & drive

Gmail, Google Calendar and Google Drive over OAuth; Apple iCloud and generic IMAP/SMTP; Apple Calendar connect. Outbound mail flows through an async queue with retries and a dead-letter queue.

●

Sandboxed execution

Code runs and URLs are browsed in isolated sandboxes. Agent safety limits and per-user, per-tenant and per-workflow cost caps prevent runaway tasks.

Governance & tenancy

●

Approval-gated actions

Reads are free; sends, deletes and shares require explicit approval. Every tool call and decision is audit-logged per tenant.

●

White-label multi-tenancy

Subdomain-resolved tenants with custom name, logo and accent colors, feature flags, privacy mode and per-tenant model-provider config.

●

Full account lifecycle

JWT auth with refresh tokens and revocation, email verification, password reset, user invites and admin controls — plus an operator dashboard.

●

Private by architecture

Tenant-scoped data isolation, prompt-injection defense on all external content, and a credential store that never hands raw provider keys to the model.

"Brief me."
↓ schedule + inbox + tasks + documents, synthesized
3 meetings today · 12 unread, 2 flagged priority · 1 overdue follow-up · contract draft awaiting your approval.

Security under the hood

The platform enforces least-privilege access end to end — your data stays inside infrastructure you control.

●

Tenant-scoped isolation

Every repository query is scoped by tenant_id. Users, documents, approvals and skills are tenant-bound — no cross-tenant leakage.

●

Credential isolation

A tool gateway mediates all provider access — agents and models never receive raw OAuth tokens or API keys.

●

Prompt-injection defense

Emails, documents, web pages and uploaded files are treated as untrusted content — tool-call authorization is independent of model output.

●

Hardened API surface

CSRF protection on mutating requests, per-route rate limits on auth endpoints, and signature or secret-path verification on inbound webhooks.

●

Encrypted data at rest

Knowledge-base source documents are encrypted at rest in Postgres, and nightly pg_dump backups keep a rolling 7-day retention.

●

Audit & observability

Every tool call and decision is audit-logged, with structured logging and OpenTelemetry tracing across the platform.

The platform

FastAPI backend, PostgreSQL 16 + pgvector, Redis 7, and a Next.js 14 frontend — Clean Architecture from domain to adapters, shipped as Docker Compose with an nginx ingress for production.

●

Model-agnostic

A model router + cost router across OpenAI and Anthropic — with per-tenant provider configuration, so each workspace picks its own models.

●

Real-time everywhere

Assistant replies, reminders and notifications push live over WebSocket — no polling, no refresh.

●

Self-host or let us host

Run the whole stack with Docker Compose in your own environment — same deployment that powers the hosted beta.

●

Extensible

A built-in tool registry with permission gating, a growing marketplace of agent packages, and optional n8n workflow automation behind a compose profile.

Get access

The beta is live at mygate.gatekali.com. Sign up with your email to create your workspace — public signup provisions a tenant with an admin account, so you get your own isolated assistant instance.

Create your workspace →