Enterprise AI, built for production

Give your AI agents a real understanding of your business.

ArcFlow Labs builds two products that work as one system. Drbool Ontology turns data scattered across databases, files, websites and APIs into business objects an agent can understand. Drbool Agent reasons over those objects, calls your tools and runs your workflows, with permissions, audit trails and observability built in.

Request a demoExplore the products
  • Private deploymentRuns inside your own network
  • Model-agnostic9 native LLM protocol adapters
  • Open interfacesREST API and Model Context Protocol
Your systems
Databases
Documents & files
Websites
REST APIs
Market & filing data
Drbool Ontology
Business objectsCompany · Contract · Ticket · Asset
Links & lineageEvery row traceable to its source
Drbool Agent
Agents & workflowsReason · Retrieve · Act
Governed actionsApprovals · Audit · Cost
Analyst session

Which suppliers had late deliveries last quarter, and which open contracts are affected?

  1. describe Supplier, Delivery, Contract
  2. aggregate late deliveries by supplier, Q3
  3. linked open contracts for top 5 suppliers

3 suppliers account for 71% of late deliveries. 12 open contracts are linked to them. Each figure links back to the query that produced it.

Illustrative example

Why it is hard

Getting AI into production is rarely blocked by the model.

Most enterprise pilots stall on the same four problems. We built our products around them.

01

Data the AI can't understand

Business meaning lives in table names, spreadsheets, PDFs and people's heads. A model that only sees raw columns gives confident, wrong answers.

02

Answers that can't act

A chatbot that can't call your ticketing, monitoring or ERP systems doesn't take work off anyone's plate.

03

No control over actions

Security teams need to know who can use what, which tools an agent may call, and when a human has to approve.

04

No visibility after launch

Without traces, cost tracking and feedback, nobody can explain a bad answer or tell whether the system is getting better.

Products

Two products. One way to put AI to work.

Each product works on its own. Together, the agent handles reasoning and action, and the ontology supplies the business context and the data behind it.

Drbool Agent

Enterprise agent platform

Build, publish, govern and observe AI applications: conversational agents, visual workflows, knowledge Q&A, natural-language analytics and single-step LLM apps, all on one platform.

  • In-house agent runtime with tool calling, sub-agents and context management
  • Visual workflow canvas with branching, loops and human approval steps
  • Retrieval engine with hybrid search, reranking and cited sources
  • Role-based access, full audit trail, cost and quality observability
Explore Drbool Agent →

Drbool Ontology

Business ontology & data layer

Connect your systems, process raw data into clean versioned datasets, and model it as business objects and relationships that any agent can query through a governed MCP endpoint.

  • Built-in connectors for databases, files, websites, APIs and public financial data
  • Visual data pipelines with document parsing and AI processing steps
  • Object types, properties and links, with full lineage back to the source
  • Analyst workspace: ask in plain language, inspect every step
Explore Drbool Ontology →

Drbool Agent

An agent platform your IT and security teams can sign off on.

Drbool Agent gives business and platform teams one workbench to create AI applications, connect them to existing systems, and publish them to the web, internal portals or any channel through an API gateway.

App Factory: five application types

  • AgentMulti-turn conversations, tool calling and sub-agent collaboration for open-ended tasks.
  • Agent FlowA visual canvas for fixed processes: conditions, loops, LLM steps, tools and human approval.
  • Knowledge (RAG)Q&A over policies, manuals, runbooks and product docs, with every answer cited.
  • NL2SQLAsk a database a question in plain language and get a validated query, results and a chart.
  • LLM appLightweight single-step reasoning, extraction and structured generation.

In-house agent runtime

Routing, prompts, memory, model calls, tool execution and auditing all run in our own agent loop rather than a third-party framework, so it can be tuned for your workloads and deployment constraints.

Production-grade retrieval

Document parsing for PDF, Office and scanned files with OCR, plus hybrid vector and keyword search, reranking, parent-child chunking and metadata filters. Answers cite their sources, and knowledge-gap reports show what is missing.

Tools and integrations

Use built-in tools, your REST APIs, MCP servers, and other platform apps as tools. Isolated code and desktop sandboxes let agents run analysis safely. Scheduled triggers and outbound notifications cover unattended jobs.

Any model, no lock-in

A multi-provider abstraction with nine native protocol adapters lets you switch or mix commercial and self-hosted models without rebuilding your applications.

Governance by default

Roles, user groups and project membership decide who can use which app and which tools. API calls, tool invocations and data changes are logged, and risky actions can wait for human confirmation.

Observability and feedback

Track requests, success rate, latency, tokens, cost, errors and app health. Drill into individual tool calls and retrieval results, and feed user ratings back into evaluation.

Drbool Ontology

Turn scattered data into business objects your agents can reason about.

Drbool Ontology runs one pipeline from raw source to governed business model: connect, process, model, serve. Every step is versioned and every value can be traced back to where it came from.

  1. 1 · Connect

    Data connection

    Sources, scheduled syncs and three kinds of data: versioned tabular datasets, file sets that keep every version of the original PDFs, Office files and web pages, and virtual tables that query the source live without copying it.

  2. 2 · Process

    Data pipelines

    Clean, parse, join and aggregate in a node-based editor with step-by-step preview. Document parsing, table extraction, chunking and AI processing are rerunnable steps with content-addressed caching. Pipelines rerun automatically when their inputs change.

  3. 3 · Model

    Ontology

    Define object types, properties and link types. Back them with managed tables fed from one or more datasets through data mappings, or with live virtual tables. Objects refresh when upstream data changes.

  4. 4 · Serve

    Agents & analysis

    Each project exposes an MCP endpoint with eight read-only tools to explore types, search, aggregate and follow links. Agents such as Drbool Agent connect with a scoped project token.

Only real changes create versions

If a sync returns the same data as last time, no new version is written and nothing downstream reruns. The dataset is simply marked as confirmed fresh. Web pages are compared on their main content, so a changed ad or script doesn't count.

Row-level provenance

Every row carries a trace back to the original API response or source file version, and that trace survives through pipelines into ontology objects. Lineage connects sources, datasets, pipelines and object types end to end.

Analyst workspace

Ask questions in plain language and follow up. Each tool step is collapsed into a readable summary with data cards, charts and tables, an outline, and a dependency graph from question to answer. Saved analyses rerun against the latest data.

Credential vault

Passwords, API keys, OAuth tokens and SSH keys are stored with envelope encryption (AES-256-GCM under a master key). Configurations hold references, never secrets.

Egress policy

Outbound requests may only reach the hosts a source allows, and private network addresses are blocked by default. Rate limits, retries and robots.txt are honored.

Simple to run

One container, one port, one PostgreSQL database. Files go to a local volume or any S3-compatible object store. Projects are isolated with their own database schema and role.

Built-in connectors

  • PostgreSQL
  • MySQL
  • SQL Server
  • Oracle
  • Server directories
  • Websites
  • Any REST API
  • SEC EDGAR
  • FRED (Federal Reserve)
  • HKEX disclosures
  • SZSE announcements
  • CNINFO
  • Tushare Pro

Any REST endpoint can be connected through the declarative HTTP connector, configured in a form with no code.

Better together

The agent thinks and acts. The ontology tells it what the business is.

Connected over the Model Context Protocol, the two products split the work cleanly, and each answer can be checked against the data behind it.

Drbool Ontology answers

  • What business objects exist and how they relate
  • Where the data comes from and how fresh it is
  • Which rows, files and requests back each value

Drbool Agent decides

  • Which questions to ask and which tools to call
  • What to compute, summarize or chart, in a sandbox
  • When to act, notify, or wait for a human

Use cases

Start with one high-value workflow. Reuse the foundation for the next.

Enterprise knowledge assistant

Employees ask about policies, manuals, SOPs and past incidents in plain language and get answers with cited sources. New hires get to expert-level answers faster.

Agent · Knowledge

Incident diagnosis for IT operations

When an alert fires, an agent gathers metrics, logs, configuration and similar past incidents, drafts a root-cause hypothesis and remediation, and routes risky actions for approval.

Agent Flow · Tools · Approvals

Ticket triage and review

Classify, enrich, check and route incoming tickets using rules, history and system data, so people spend their time only on the exceptions.

Agent Flow · Knowledge

Self-service analytics

Business users ask "average resolution time by team this month" and get a validated query, the result and a chart, without waiting on a data team.

NL2SQL · Ontology

Financial and market research

Sync SEC filings, FRED economic series and exchange disclosures, parse the documents, model companies and filings as objects, and let analysts question them with full provenance.

Ontology · Analyst

Scheduled checks and reports

Agents run on a schedule to check system health, query metrics, summarize anomalies and deliver daily or weekly reports to your team's channels.

Agent · Schedules · Notifications

Security & trust

Every action is visible, controlled and accounted for.

Governance isn't an add-on. Permissions, audit and isolation are part of the core data model of both products.

  • Private deploymentInstall both products in your own data center or cloud account. Your data stays in your environment.
  • Role-based access controlRoles combine with project membership to decide who can see, build, run and publish.
  • Human in the loopHigh-risk tools and workflow steps can require explicit approval before they run.
  • Full audit trailAPI access, tool calls and data changes are recorded and can be reviewed.
  • Project isolationEach ontology project has its own database schema and role. Data sources can be shared across projects only by their owners.
  • Encrypted secretsEnvelope-encrypted credential vault, with configurations holding references instead of secrets.
  • Network egress controlOutbound requests are limited to approved hosts, with private addresses blocked by default.
  • Verifiable releasesVersioned offline installers with SHA-256 checksums, and documented upgrade paths for every release.

Deployment & engagement

From first workflow to company-wide platform.

  1. 1

    Discovery

    We work with your team to pick one workflow where the value is clear and the data is reachable.

  2. 2

    Proof of concept

    We deploy in your environment, connect real sources, model the core business objects and build the first agent against your own data.

  3. 3

    Production

    We harden permissions, approvals, monitoring and integrations, then roll out to users through the web, internal portals or the API.

  4. 4

    Scale

    Your team adds new sources, objects and applications on the same foundation, with our support and regular product releases.

Company

About ArcFlow Labs

ArcFlow Labs builds software that helps enterprises move AI from pilot to production. We believe the hard problems are less about which model you use and more about context, control and accountability: agents need to understand the business, act only within clear boundaries, and leave a record anyone can check.

Our two products, Drbool Agent and Drbool Ontology, are built in-house from the runtime up. We ship them as self-hosted software, so the data, models and processes stay under your control.

Focus
Enterprise AI agents and business ontology
Products
Drbool Agent · Drbool Ontology
Delivery
Self-hosted, private cloud or managed

Get in touch

See it working on your own data.

Tell us about the workflow you want to automate. We'll set up a tailored demo or a proof of concept in your environment.

Or email us at contact@arcflowlabs.ai