Egidia The agentic business OS

Run your business on agents you actually control.

A team of agents that take action, remember everything across sessions, and meet to decide together. Running on your infrastructure, inside your boundary, with a signed record of every move.

Live · workspace T1 · T2 · T3
PAPersonal Assistant
OpsOperations
CommsComms
FinFinance
Awaiting approval, T3 ⌘ ↵
Deploy
Self-hosted Docker, customer cloud, or managed EU SaaS.
Model layer
Hot-swap providers per agent. Local Ollama supported.
Receipts
Signed audit log per workspace. Verifiable offline.
Jurisdiction
Data stays in the EU region you nominate. Full stop.
01What it does

A team. Not a chat window.

Most "AI" tools draft text and wait. Egidia runs the work; agents take action, hold context across sessions, and meet to decide together when something needs more than one head.

01 / a They act, not advise Policy-gated actions
01 / b They remember Three tiers, cross-session
01 / c They meet, decide, log Multi-agent room
Show what they do, in three
01 / 03

They act, not advise.

Triage the inbox. Update the project tracker. Reconcile the books. Draft the brief. Every action runs against a policy you set; every action lands in the audit log.

02 / 03

They remember, at every level.

Tiered memory (each agent's own, the team's shared context, the organisation's institutional record). They know your clients, your decisions, your last quarter's calls.

03 / 03

They meet, decide, log.

The Meeting Room: a multi-agent room with templates (strategic decisions, incident response, board prep). Agents propose, debate, vote. Rationale is sealed in the record.

02The team

A founding team that shows up to work.

Your Founder Team is your Personal Assistant plus 3 Agents you pick from a roster of 7. Choose a preset built for your shape of business — Solo Founder, Micro-Agency, or Tech Startup — or build your own. Add more Agents from the store as you grow.

Showing preset
Solo Founder Cut 25h/wk of ops, comms, and money admin while you build.
PA · Always included

Personal Assistant

Owns your inbox and calendar.
  • Triages email by sender, intent, urgency
  • Drafts replies in your voice
  • Books meetings against real availability
  • Surfaces what needs you; mutes what doesn't
Anchor · T1
Slot 1 · Operations

Operations

Keeps work moving across the team.
  • Triages tasks across owners
  • Unblocks projects when they stall
  • Drafts standup and weekly notes
  • Flags risks before they slip
Swap anytime
Slot 2 · Comms

Comms

Writes what the business says.
  • Drafts posts in your voice
  • Handles social replies and DMs
  • Press lines and one-pagers
  • Internal updates and FAQs
Swap anytime
Slot 3 · Finance

Finance

Watches the money.
  • Reconciles books weekly
  • Flags burn anomalies early
  • Drafts investor updates
  • Expense triage and receipts
Swap anytime
Specialists

Strategy, Engineering, and more.

Roles you'd hire when the founding team isn't enough. Add per workspace; retire when the project closes.

Industry packs

Domain-shaped operators.

Operator packs trained for specific industries. Rolling out from Phase 2 as the catalog expands.

Custom agents

Your knowledge; our runtime.

Define the persona, the skills, the routing, the approval tier. A guided wizard turns it into a working operator on the platform.

03Meeting Room

The room where agents decide together.

Open a template, drop in the question, walk away. A team-leader agent runs the agenda; specialists weigh in. You arrive to a recommendation with the reasoning attached, and a button to approve it.

03Open the five meeting templates
03 / templates

Each template fixes the agenda, the roles, the approval gate, and the artefact at the end. Sessions are written to the workspace; rationale is sealed in the audit log.

Party ModeOpen-ended brainstorm; every agent contributes.
Strategic DecisionA proposal, three perspectives, one recommendation.
Incident ResponseTriage, contain, communicate, document; timeline included.
Compliance-First 1.5Produces a compliance matrix per your industry.
Board Prep 1.5Quarter synthesised into board-ready slides + Q&A bank.
04Tiered memory

They remember, like a team that's been here a while.

Memory at three levels, modelled on how real teams hold knowledge. Cross-session, queryable, bounded by the same policies that gate actions. The thing that turns "AI tool" into "colleague".

Tier I · Agent

Each agent's own.

What it learned this session, what worked, what to try next time. Persists across runs, not just the current window.

Tier II · Team

Shared in the workspace.

Project history, client relationships, the decision on the Acme retainer in March. Any agent on the team can read it; the team-leader curates it.

Tier III · Organisation

Institutional record.

Policies, brand voice, what the company has and hasn't done. Read by default; write by approval. The thing a new hire would take six months to learn.

05Governance-first

The audit trail your regulator already speaks.

Approvals, sanitisation, signed receipts. The plumbing a CISO or DPO opens before approving a pilot. We wrote it for them.

05 / a Three approval tiers T1 routine → T3 sync
05 / b Signed audit log SHA-256 chained
05 / c Security stack 5 layers, opt-in
05Open the full governance model
05 / a

Three approval tiers. One rule.

Every action runs at the lowest tier that fits. Routine work happens in the background. Anything that touches state or the outside world surfaces for human review, with the context bundle already attached.

T1 Routine work; runs and logs. No human
T2 Internal state change; owner notified, short veto window. Async human
T3 External or material action; explicit approval required. Sync human
05 / b

Signed audit log per workspace.

Every agent action and every human override is sealed with a SHA-256 hash chained to the previous entry. Exportable as JSON, verifiable offline. The format your DPO already speaks.

14:02:11agent.contract-reviewer read /clients/acme/MSA.pdf9f3a…7b91
14:02:14agent.contract-reviewer proposed redline §4.28e12…cc40
14:02:32human.paul approved redline · T2d771…0a5c
14:02:33agent.contract-reviewer wrote /clients/acme/MSA-v2.pdf2bc8…1f04
Presidio Data Sanitisation Proxy; redacts PII before it reaches a model. Default · basic
RuneLM Advanced sanitisation for regulated workloads; stacks on top of Presidio. Advanced · opt-in
BonkLM LLM-Guardrails; catches prompt-injection, jailbreaks, hallucinated tool calls. Advanced · opt-in
Hallucination Detector Scores every model output for factual confidence; configurable threshold per workspace, failed outputs trigger re-roll or human review. Advanced · opt-in
AI Shadow Detection Watches the workspace for unsanctioned AI traffic, rogue model calls, and shadow agents spun up outside policy; flags them in the audit log. Advanced · opt-in
06Hot-pluggable

Swap every layer. No restarts.

Every layer is a plugin. Switch the model behind an agent at lunch, add a connector at dinner, replace the storage backend without taking the runtime down.

06 / a Models 6+ providers
06 / b Connectors 14+ tools, growing
06 / c Platform MCP MCP-native
06 / d Agent Store Packs + custom wizard
06 / e OSS SDK 3 open kits
06Open all five plugin layers
LLM providers

Multiple providers ship by default; add what you need.

Top-tier defaults cover most agents. Drop in extras from the registry as plugins. Local Ollama for fully offline workloads.

Anthropic OpenAI Mistral Google DeepSeek Ollama · local + more via plugin
Connectors

An extensive catalogue; growing on every release.

The tools a working business actually uses, wired in. New connectors land when a customer asks, not on a roadmap chart.

Google Workspace Microsoft 365 Gmail Outlook ProtonMail Slack Notion Stripe GitHub Linear HubSpot Asana Google Drive Calendar and growing
Platform MCP

Manage the platform itself from anywhere.

Egidia ships its own MCP server. Spin up agents, dispatch tasks, approve actions, and pull audit data from your IDE, your terminal, or any MCP-aware client. No bespoke API to learn.

Agents Workspaces Approvals Audit log Memory MCP-native
Agent store

Founder Team, Specialists, or your own.

Default Founder Team in every install. Specialists pack rolls out Phase 1.5. A guided wizard for custom agents (your domain knowledge, our runtime contract).

Founder Team Specialists Custom agent wizard
OSS SDK

Open-source kits for fractional CTOs and power users.

The platform is closed-source; the SDK is not. Build your own agents and connectors against the same contract our defaults use.

@egidia/agents-kit @egidia/plugin-sdk @egidia/connector-kit
07Sovereignty

Nothing leaves your boundary unless you say so, and when it does, it's signed.

07 / a EU residency Region you nominate
07 / b Local Ollama In-boundary by default
07 / c Self-host or sovereign Your box or your cloud
07 / d Sub-processors Named in your DPA
07Open the full sovereignty model
Data residencyEU region you nominate. Sovereign infrastructure on Phalanx Sovereign and Citadel.
Model layerLocal Ollama supported. Cloud models opt-in per agent, per workspace.
InfrastructureSelf-host on your box, or sovereign deploy in your cloud.
Sub-processorsEvery third party that touches your data is listed in your DPA by name and by what they touch. We never add one without telling you first.
SOC 2RoadmapTargeting Q3 2026.
Your boundary
Agent runtimein-boundary
Memory layerin-boundary
Audit login-boundary · signed
Storagein-boundary
Model · local Ollamain-boundary
Optional · per agent: outbound call to a cloud model, sanitised by Presidio, logged in the audit trail.

A small early-access cohort. Claim your spot.

A twenty-minute call. We ask three questions, you ask three. If Egidia fits, you're in. If it doesn't, we'll tell you who does.