Abstract geometric composition with teal and navy tones representing AI-powered executive workflows
Beta Available Now

Run your own AI
executive assistant.
Keep control.

Geoffrey helps with tasks, meetings, docs, follow-ups, and research — while keeping your data local and actions under your control.

See how it works
PromptShield ON by default
Draft Queue — nothing sends without approval
Verified outputs with proof bundles
Open-source release planned
Scroll
The Problem

Most AI assistants are magic without control.

Geoffrey flips the script. You get the productivity gains of an AI executive assistant — without the chaos, the cloud lock-in, or the trust deficit.

Common agent failures

Unreliable

Agents hallucinate, miss steps, and produce inconsistent results.

Unsafe

Unchecked access to email, calendar, and files with no guardrails.

Expensive & noisy

Token-hungry, notification-heavy, and hard to justify the cost.

Hard to operate

Complex setup, cloud dependencies, and opaque decision-making.

Geoffrey's approach

Deterministic task plans with verified skills — not prompt-and-pray.

PromptShield, approval queues, and audit trails built in from day one.

Self-hosted with local models. Your hardware, your budget, your data.

Docker pull, setup wizard, ready in minutes. No cloud account required.

Abstract shapes illustrating GeoffreyBot capability pillars
Capabilities

Six pillars. One assistant.

From daily tasks to multi-step research projects with professional deliverables — Geoffrey handles the full workflow, not just the first turn.

Personal Assistant Core

Natural language tasks, daily summaries, meeting prep with notes and action items. Quick answers, clear next actions — like a real assistant in chat.

  • Tasks & reminders with due dates
  • "Today / What's next?" daily views
  • Meeting notes + follow-up tasks

Agentic Workflows

Multi-step projects orchestrated end-to-end. Swarm roles handle complex tasks in parallel — researcher, analyst, writer, reviewer — with budgets and progress tracking.

  • Projects with pause/resume/cancel
  • Background jobs + notify-on-done
  • Mission Control with schedules & triggers

Web Research & Intelligence

Guarded web access searches the public internet by default with SSRF protections. Grounded outputs cite sources and timestamps — no cutoff-style guessing.

  • Auto-activation wizard for first use
  • Evidence highlights + sources used
  • Partial results with failure reason codes

Deliverables & Output Quality

Professional outputs you can send to investors or clients immediately. PDF reports with citations, XLSX financial models, social content packs — all in one run.

  • Artifact Factory (PDF, XLSX, social)
  • Bundle exports with SHA256 hashes
  • One-click templates: OSINT, investor memo, market map

Safety & Governance

Draft Queue ensures nothing executes without explicit approval. Omni-channel approvals, network policy controls, claim verification, and tamper-evident proof bundles.

  • Human-in-the-loop execution by default
  • Verification layer (pass/warn/fail)
  • Trust receipts + proof bundles

Integrations & Interop

Telegram, Dashboard Chat, and optional connectors for WhatsApp, Slack, and Discord. A2A agent interoperability and MCP tool server support built in.

  • OpenClaw overlay integration
  • Inbound/outbound agent calls
  • Secure MCP tool connections
Safety & Governance

Institutional-grade controls, not vague promises.

We don't claim Geoffrey is "safe" in the abstract. We ship specific, auditable mechanisms that let you verify safety for yourself. Nine layers of control — from network policy to cryptographic proofs.

PromptShield

Untrusted content is automatically flagged and sandboxed. Prompt injection attempts are caught before they reach the execution layer.

Draft Queue & Approvals

Human-in-the-loop execution by default. Nothing is sent, posted, or executed externally without explicit approval. Omni-channel approve/reject from Telegram or WhatsApp with replay-protected tokens.

Network Policy Controls

One dashboard to set how strict web access should be: deny, ask, allow with allowlist, or guarded mode. Configurable caps on request size, timeouts, and redirects — no config files required.

Verified Skills

Skills are scanned at install time. Unverified skills are quarantined with a kill switch. You control what Geoffrey can do.

Verification Layer

Claims in reports are graded against evidence: pass, warn, or fail. Reduces hallucination risk and increases confidence for business use.

Trust Receipts & Proof Bundles

Tamper-evident audit chains and exportable proofs. Outputs can be validated and shared with cryptographic-style attestations.

Pairing Moderation

Approve, deny, or block access requests from the dashboard. Prevents unknown users from gaining access to your instance.

External Malware Scan

Optional malware scanning lane with BYO API key. Off by default — you opt in only if your threat model requires it.

Doctor Posture Checks

Automated health checks validate Geoffrey's configuration, permissions, and connectivity. Issues surface before they become problems.

Beta Scope
"Trust first. Automation second."

Geoffrey suggests and executes with your approval. Nothing happens externally without explicit sign-off. We earn trust before we ask for autonomy.

Live in beta

  • Personal assistant: tasks, daily views, meeting prep
  • Projects & multi-step workflow orchestration
  • Web research with Guarded mode and evidence trails
  • Artifact Factory: PDF reports, XLSX models, social packs
  • Template library: OSINT brief, investor memo, market map
  • Background jobs with progress and notifications
  • Draft Queue for all outbound actions
  • Gmail summaries and calendar agenda prep
  • Verification layer with claim grading

Not in beta yet

  • Auto-send email on your behalf
  • Create, update, or cancel calendar events
  • Unattended external posting by default
  • OpenClaw Bridge integration
  • Full Mission Control scheduling
  • Write access to any connected service

These capabilities are on the roadmap. We'll ship them when the safety infrastructure is ready — not before.

Getting Started

From zero to Ready in minutes.

No sales calls, no onboarding sessions, no waiting lists. Three steps and you're running your own AI executive assistant.

Three-step process: Download, Configure, Launch
01

Register & Download

Sign up for the beta, get your access link, and pull the Docker image. No cloud account, no API keys to configure upfront.

02

Complete Setup Wizard

The guided wizard walks you through connecting your tools, setting policies, and configuring preferences. Status: Ready.

03

Launch a Project or Pick a Template

Start a multi-step project, pick a one-click template — OSINT brief, investor memo, market map — or just ask a question. Results in minutes, not days.

FAQ

Questions answered.

Straight answers to the questions we hear most. If yours isn't here, reach out — we're happy to talk.

Register for the GeoffreyBot beta and get to Ready in minutes.

Limited spots available. Read-only external connectors in beta.