Some follow-ups never sent. Some contract sections signed without anyone noticing. Some obligations aging out without a clear owner. Some commitments drifting away from stated strategy. Each miss looks small. Compounded across a year, it translates to lost revenue, missed renewals, partner-relationship damage, or compliance exposure.
AEGIS is the vigilance layer that catches it — across CRM, mail, documents and calendar — and proves each catch with evidence before it costs you. A builder can't grade its own work, which is why AEGIS watches beside the systems you already run, not inside them.
The record you hold is the Catch Ledger: what we caught, with evidence, before it bled out.
AEGIS runs continuously on its own operating tenant — what you see below is the loop catching real follow-ups, partner openings, and strategic postures that would otherwise have aged out silently. Strategic Memoria nodes accepted on the support tenant stand as live evidence today.
"Follow up with an enterprise prospect regarding an AI solution discussion routed through a senior executive in the same group." Surfaced from call notes, accepted Memoria strategy, and dormant correspondence. Signals across multiple systems point to one un-owned follow-up.
"A partner candidate indicated sponsorship interest — schedule outreach before the window closes." Caught from cross-system signals: partner correspondence, calendar gaps, and accepted strategy about partner posture. The window would have closed silently.
"Strategic posture: engage personal and family network for organizational access into target accounts." A partner_strategy node — a durable operating posture extracted as memory, reusable as context for the next conversation. Without this layer, that posture lived only in someone's head.
None of these needed AI to invent them. They needed a vigilance layer to notice them. The catch was the product. Every accepted catch is now durable organizational memory — provenanced, auditable, reusable. The running record you hold is the Catch Ledger: what we caught, with evidence, before it bled out.
The vigilance loop runs continuously across the seams modern enterprise tools don't watch. It distinguishes evidence from accepted memory, asks for human approval on consequential writes, and never autonomously executes.
Across CRM, email, documents, calendar, tasks, notes. Notices when commitments go unclosed, when handoffs drop, when behavior drifts from stated strategy, when silent damage accumulates against a healthy baseline.
Anything retrievable in your knowledge base, vault, or drive is evidence. What your company has approved as true — with provenance, confidence, and supersession trail — is accepted memory. The two are never confused.
AEGIS detects, analyzes, drafts, and proposes. It never autonomously sends an email, updates a CRM stage, amends a contract, or books a calendar slot. The accept gate is the trust boundary.
Each accepted catch becomes a Memoria node — tenant-scoped application data with provenance, trust labels, and source references. Not model memory. Not fine-tuning. Reusable as context for the next workflow.
Permanent refusal: AEGIS never autonomously executes a consequential write — no email, no CRM update, no contract amendment, no calendar booking — without explicit human approval. This is a Founding Contract clause, not a roadmap item.
A builder cannot grade its own work. The platform selling an agent — and the vendor deploying it — are structurally disqualified from being the independent check. AEGIS is a fabric of governed agents built to be that check: each does scoped work under a contract, sees only what it needs, cannot quietly overspend, never writes memory alone, and leaves a trace for everything. That is precisely what lets AEGIS stand beside the systems you already run, not inside them.
Each agent runs under a contract with least-privilege over the layers and sources it may touch — never broader than its job.
A per-agent monthly budget hard-gates that agent's tasks once reached; per-tenant spend is metered, attributed, and alerted.
No conclusion becomes action or memory until a human approves it, and every action is logged for the record.
Recurring work becomes a reusable skill, so agents call the model less as they mature — lower cost, higher reliability.
A public schematic for IT review: inputs stay tenant-scoped, agents operate under policy, and only approved catches become durable memory.
CRM, mailbox, documents, calendar, tasks, Teams and M365 history enter as evidence.
Schema-per-tenant PostgreSQL, RBAC, audit logs and per-source permissions hold the data line.
Channel events are recorded with source identifiers before work begins, so retries do not become duplicate actions or silent loss.
Omni routes work to the specialist roster under tier-aware model policy and least privilege.
AEGIS surfaces the missed follow-up, clause risk or stalled commitment with source evidence.
Consequential writes and durable memory wait for a person. No silent autonomous execution.
Accepted catches become trusted context; recurring patterns harden into reusable skills.
For deeper IT diligence, use the Evaluator Q&A, Enterprise IT runbook and Tech Stack. The private schematic can be reviewed under NDA.
AEGIS is not another place users must remember to check. The same Catch Ledger, approvals, and audit trail surface through the channel that fits the moment.
Windows 10/11 pilot MSI validated: tray badge, notifications, quiet/resume, start-on-login, and direct entry to Decisions. Signing and auto-update remain hardening work.
/omni/standalone is the compact operator surface for decisions, catches, actions, source status, and chat with the governed agent team.
/os and /desktop-v2 remain the deep work surfaces for CRM, Catch Ledger, Memoria, Boardroom, workflows, settings, and audit context.
Operational channels can receive catches, read commands, digests, and approval prompts while sensitive writes still pass through the human gate.
AEGIS is live in production today. Each item is tagged so diligence can separate shipped product, configured/pilot-gated capability, and the next build priorities.
What AEGIS will not become: a universal enterprise agent mesh, autonomous executor, or system that fine-tunes on tenant data. These refusals are permanent.
A pilot is narrow, measurable, and governed. No step obligates the next. Commitment grows only as evidence accumulates.
Load public themes and product context. Run Opportunity Radar against CRM, email, Memoria, tasks, calendar. Save useful catches to Memoria. Re-run context-aware output. Evaluate quality through dogfood scoring. Turn the pattern into a repeatable Skill. Directly demonstrates the Intelligence Loop on a workflow the buying team already runs.
Controlled layer for contract, policy, or transformation document review. Select source. Choose or skip policy comparison. Run analysis. Draft corrections. Approve selected outputs. Save to Memoria. Export approved artifact. Run prior-review or delta path on a later version. Simple to understand for legal, compliance, IT, and business audiences.
Use AEGIS around a transformation program where decisions, risks, and stakeholder communication must persist. Upload artifacts. Ask for current state and unresolved risks. Run executive weekly briefing Skill. Capture approved decisions into Memoria. Generate stakeholder-specific brief. Inspect provenance and prior decisions.
Typical pilot shape: 60 days, one domain, controlled data scope, 10–25 users, no autonomous external writes in v1, explicit approval gates, pre/post scoring against baseline, final report on value, governance, reuse, and roadmap.
Vigilance is the product. Sovereignty is how it ships safely. The IT-side guarantees that make pilots viable in regulated, sovereign, or on-premise environments.
Your PostgreSQL. Your object store. Your audit log. AEGIS never writes to a shared cloud database. Every tenant runs in an isolated schema — and in Enterprise deployments, that schema lives on your own servers.
SAML 2.0 with Azure AD / Entra ID means employees can log in with credentials your IT team already manages. No new password universe. Provisioning and deprovisioning remain anchored in your IdP.
Microsoft Teams connector path with Azure Bot setup. Employees can reach AEGIS agents from Teams while the portal remains the control plane for approvals, evidence, memory, and audit.
Identity, channel, access, audit, and hosting controls are implemented or configuration-ready. Each deployment walks through the exact customer environment, provider policy, and data boundary before go-live.
Agents can appear as a bot inside Teams after Azure Bot setup. Employees message AEGIS from a channel or DM while agent intelligence, KB context, CRM data, and Memoria remain governed by the tenant's AEGIS backend.
Setup guide →Configuration path for Azure AD / Entra ID, Okta, Google Workspace / Cloud Identity, OneLogin, Ping Identity, JumpCloud, Keycloak, and ADFS — any SAML 2.0-compliant provider. IT connects Federation Metadata XML and AEGIS joins the customer identity layer. Native OAuth/OIDC tiles for Okta and Google coming soon.
Setup guide →Assign Admin, Manager, or Member roles per user. Managers cannot touch billing or agent configuration. Members cannot see other teams' data. Roles enforced at the API layer — not just the UI.
Roles reference →Agent actions, login events, config changes, and API calls are logged with timestamp, user, outcome, and token cost where available. Export to CSV from Settings → Security for GDPR, ISO 27001, and internal audit workflows.
Audit reference →The AEGIS stack — PostgreSQL, LiteLLM, the web app, nginx — ships as a Docker Compose path for Linux hosts in a datacenter or private cloud. Customer-specific networking, identity, model routing, and backup policies are configured during deployment.
Deployment runbook →For restricted or air-gapped environments, AEGIS can route inference through a local Ollama instance. Supported model families include Llama, Qwen, Mistral, and GGUF-compatible models the customer already operates.
LLM routing docs →AEGIS complements the CRM instead of replacing it: native AEGIS CRM is live, and HubSpot is connected as a source-aware CRM proof with OAuth-backed reads, Omni/Telegram pipeline and follow-up/attention views, Radar/Catch Ledger evidence, and provider links. Explicit import/write-back, webhook hardening, and additional CRM adapters are prioritised by pilot demand.
OAuth connection for contacts, companies and deals as read-through external evidence. AEGIS can answer HubSpot-qualified follow-up and attention questions, then surface opportunities in Omni, Telegram, Radar and the Catch Ledger with direct provider links.
Planned Sales Cloud adapter for leads, opportunities, accounts, and pilot-scoped custom objects.
Notify me when live →Planned Dynamics adapter paired with Teams, with write actions gated by customer policy.
Notify me when live →Planned adapter for pipeline, contacts, and deals, sequenced by customer demand.
Notify me when live →Planned adapter for deals, contacts, and activity logging with human-approved write-back.
Notify me when live →Missing your CRM? Tell us — integrations are prioritised by demand.
Copilots are useful inside their suites. AEGIS is designed for the seams between systems: the missed follow-up, stale commitment, unowned obligation, approval trail, and accepted memory.
| Capability | AEGIS Enterprise | Microsoft Copilot / Suite Agents | Generic AI Tools |
|---|---|---|---|
| Primary job | Independent vigilance layer across operating seams | Productivity and task assistance inside Microsoft ecosystem | Prompted generation, search, summarisation |
| Cross-system seam watching | CRM + mailbox + documents + calendar + tasks + memory | Strongest inside M365/D365 and connected Microsoft Graph sources | Manual context assembly |
| Catch ledger | Proposed catch → evidence → approve/reject → accepted memory | Activity and audit surfaces inside suite workflows | Usually none |
| Human approval before durable memory | Explicit gate for consequential writes and Memoria promotion | Depends on app, workflow, and tenant configuration | Prompt-level only |
| Data residency | Managed EU / BYOC / customer DB path | Microsoft tenant and regional controls | Vendor cloud |
| Private or self-hosted deployment | Private deployment path; self-hosted annual license for customer-funded inference | Cloud service inside Microsoft estate | Rare |
| SAML / Entra / Teams | SAML path + Teams connector via Azure Bot setup | Native first-party strength | Often capped |
| Cost model | Plan + soft task notifications + premium-AI caps; self-hosted annual SKU | Seat/licence plus credits or suite consumption | Token/message based, often variable |
| Audit export | Audit trail + CSV export path; model-policy records live | Purview and Microsoft admin/audit surfaces | Varies |
| CRM integration | Native CRM live · HubSpot read/evidence + follow-up routing live · broader CRM sync roadmap | Dynamics 365 native | Varies |
| Architecture transparency | Public tech stack + IT runbook + evaluator Q&A + NDA schematic | Vendor documentation and tenant admin center | Usually limited |
Pick the model that matches your security posture. You can migrate between modes without data loss.
We host on EU-based infrastructure. Your data is isolated in a dedicated PostgreSQL schema — never shared, never used for training. Up in 24 hours.
We ship the Docker Compose stack. Your team runs it on AWS, Azure, GCP, or Hetzner. We handle updates; you control the perimeter and the data.
Full local stack with local model routing. Designed for restricted outbound environments in regulated industries: banking, defence, healthcare, government.
All three modes use the same 8 agents, the same UI, the same APIs, and the same audit trail. You're not choosing a different product — you're choosing where it runs.
Beyond the refusals AEGIS has made permanent (no autonomous consequential writes, no fine-tuning on tenant data, no universal-mesh pretensions), these commitments hold for partner-mediated and regulated-customer deployments.
Tenants declare region. Cross-region processing requires explicit per-occurrence approval and is logged. Regulated plans receive a per-region deployment guarantee.
Publicly listed registry of every sub-processor and model provider. Tenants notified before changes. Regulated plans may veto specific providers; the model gateway routes around them without service interruption.
Evidence, Memoria graph with edges and provenance, entity graph, audit log, tenant configuration — extractable in standard formats at any time. No lock-in by design.
Per-tenant retention windows. Legal hold suspends all deletion paths — including Guardian retention sweeps — until released. Scope by entity, time window, or memory branch.
When an upstream provider deprecates a model AEGIS depends on, tenants are notified at least 30 days before AEGIS-side end-of-life with a documented migration path. Tenants may pin specific model versions for compliance-sensitive Skills.
When a tenant grants delegated access to a partner during a transformation engagement, the scope of partner authority, the audit trail of partner actions, and the provenance of partner-introduced memory are all explicit, exportable, and revocable.
These commitments are part of the AEGIS Founding Contract §IVA and bind every deployment, demo, and pilot.
A 30-minute call is enough to walk one live catch on AEGIS's own operating tenant and scope a controlled pilot for yours.
Or email directly: [email protected]
