Documents, connectors, Company-DB, and agents in one stack

The operating system for company data and agent work

AI CEO ingests financial, legal, tax, governance, operational, asset, and general business data, promotes verified outputs into Company-DB, and gives agents and operators a compact way to search, reason, approve, export, and act without losing evidence or control.

Multi-company tenancy
Approval-gated agent actions
Git-backed Company-DB history
Dashboard
Executive Overview
Finance, document pipeline, and company signals in one view
Revenue
$9.6M
Expenses
$5.0M
Net Profit
$4.6M
Financial trend
Latest verified period: 2025-12
Verified
Gross margin
64%
Opex ratio
15%
Cash delta
+$1.29M
Jun
Jul
Aug
Sep
Oct
Nov
Dec
Document questions
Bank statement
Pending
Need currency + account owner
Lease transfer
Review
Manual legal review
Payroll export
Done
Mapped and promoted

Company-DB source of truth

Verified records, compact summaries, write queues, semantic sidecars, and drill-down paths all stay behind one contract surface.

Agent-first runtime

Chat, MCP, connector actions, documents, artifacts, and approvals are designed for agents first, not bolted on later.

What the platform actually includes

This is not just a chat UI. It is the full stack around company data: ingress, routing, verification, storage, retrieval, operator workflows, and production controls.

Document ingestion

Upload PDFs, spreadsheets, statements, contracts, policies, and operational files. Keep raw evidence, clarification loops, and processing history together.

Connectors and live systems

Work with Google Drive, Slack, Jira, BambooHR, Dynamics BC, Payhawk, Zendesk, Telegram, Microsoft, and other bounded connector surfaces.

Company-DB

Promote decision-grade outputs into a git-backed Company-DB with write queues, summaries, RBAC, and a stable retrieval contract for agents and operators.

Agent workflows

Run internal agents and consultant chat on compact, auditable context instead of giant raw payloads. Keep actions approval-gated when the outcome matters.

Ops and observability

Track queues, review flags, clarification requests, promotions, semantic canaries, and admin drill-downs without losing tenant boundaries.

One platform across every domain

AI CEO is built for cross-domain company work. Finance does not live apart from legal, operations, knowledge, or communications.

Finance
Banking
Revenue
Expenses
Legal
Tax
Governance
Operations
Assets
Communications
Knowledge
People
Documents

Retrieval that matches the data

The platform does not force every question through one generic search path. Finance, documents, and narrative material each keep the retrieval mode that preserves quality.

Structured finance retrieval

Financial statements, ledgers, bank imports, and snapshots stay structured-first. Finance is not flattened into generic semantic search.

Summary-first Company-DB reads

Agents and users hit compact summaries first, then drill into verified files, records, or raw evidence only when the question really needs it.

Semantic narrative search

Narrative domains like legal and knowledge can use semantic sidecars behind Company-DB policy enforcement without replacing the source of truth.

Retrieval
Search without flattening everything
Structured finance, summary-first Company-DB, and narrative search can coexist
Finance
Company-DB
Semantic canary
Search: lease transfer payment obligations
legal/imports/lease-transfer.qmd
Summary-first hit
Clause, payment obligations, renewal terms
legal/_summary.qmd
Semantic fallback
Red flag register, unresolved obligations
finance/_summary.qmd
Structured finance read
Verified latest period and expense estimate

How information moves through the system

From upload or connector event to verified answer, each step is explicit and auditable.

01

Ingest from every entrypoint

Uploads, connectors, imports, webhooks, and background jobs all enter through bounded ingestion paths.

02

Normalize by document class

Narrative documents become markdown-like artifacts, table-heavy or financial data stays structured, and OCR-heavy files stay on hard paths.

03

Promote through write queues

Nothing becomes authoritative by accident. Company-DB writes stay auditable, replay-safe, and tied back to source evidence.

04

Retrieve the right way

Finance uses exact and structured access, narrative documents use compact search and semantic fallback, and raw material stays drill-down only.

05

Act with approvals and exports

Agents can draft artifacts, create exports, ask clarifying questions, and request approval before consequential actions.

Connectors, imports, and live agent actions

Some systems sync into Company-DB, some are used live in bounded reads, and both paths are first-class parts of the product.

Connect and sync

Google Drive document import and watched folders
Email and external file ingress
Financial document parsing and promotion
Background jobs for retries, reviews, and queue repair

Use live systems in-place

Slack and Jira hub actions for operational reads
BambooHR, Dynamics BC, Payhawk, Zendesk, Microsoft, Telegram, and Odoo connector actions
Bounded pagination, explicit counts, and compact results so agents do not hallucinate system-wide truth from one page
Connectors
Sync where it matters, read live where it helps
Operational systems do not all need the same ingestion mode
Google Drive
154 docs
Syncing
Slack
182 channels
Live read
Jira
Projects
Live read
Payhawk
Fund accounts
Live read
Dynamics BC
Invoices + GL
Live read
Zendesk
Tickets
Live read
Recent connector activity
Drive import
2m ago
5 files promoted
Slack read
live
engineering, random, design
Payhawk
live
3 fund accounts returned
Dynamics BC
live
14 companies discovered

Built for agent delivery, not just operator clicks

The system includes the machine-facing layer too: scoped API access, MCP, thread workspaces, artifacts, exports, approvals, and explicit publish paths into shared company knowledge.

Scoped external agent access

Use bearer API keys with company-scoped access policies and granular scopes for connectors, Company-DB, documents, people, settings, and dashboard summaries.

Personal threads on company context

Each operator gets personal chat threads that can reason over the current company while keeping thread state, attachments, artifacts, and approvals isolated.

App-level MCP and control plane

Expose one machine-friendly surface for connector actions, Company-DB reads, document actions, people records, and company control operations.

Artifact and export pipeline

Agents can draft memos, create real spreadsheet exports, store them in thread workspaces, and publish them to the shared document layer only when the user decides to.

Agent
Thread, artifact, approval, publish
Personal workspaces and shared company knowledge stay separate until you decide otherwise
Review the uploaded lease and draft a negotiation memo. If it looks usable, prepare a company document draft.
I found the lease import, highlighted payment obligations, and created a draft memo artifact. Publishing it to company documents still needs approval.
Artifact draft
Pending approval
lease-negotiation-memo.qmd
Personal thread artifact with draft summary, obligations, renewal clauses, and open legal questions.
Approval path
Draft created
Done
Approval requested
Pending
Publish to company docs
Blocked

How agent work gets delivered safely

Personal thread artifacts stay private until explicitly shared to company documents
Approval requests gate consequential writes and publish actions
External machine clients can use REST and MCP without browser-only setup flows
Company-level context, connectors, and Company-DB remain available behind one contract
Use the same platform for internal operators, embedded chat, external machine clients, and company-scoped agent workflows without splitting storage, approvals, or retrieval logic across separate products.

Built to stay controlled under real operations

The product is designed for production teams that need agents, but also need boundaries, review paths, and clean auditability.

Tenant-scoped access

Every company stays isolated. Roles, memberships, and company slugs define what can be read or changed.

Audit trails and history

Document reviews, promotions, Company-DB commits, and background jobs leave a trace that operators can inspect.

Clarifications before bad data

When a document is incomplete or ambiguous, the system asks questions instead of silently guessing.

Approval-gated actions

Exports, consequential writes, and consultant actions can stop for approval instead of executing optimistically.

Full product surface, not just a chatbot

Use one system for company data, operators, and agents

Bring in documents and connected systems, normalize them safely, store verified outcomes in Company-DB, and let agents work on compact, decision-grade context with approvals, exports, and drill-downs when needed.