Grounded AI agents
Agents answer from your documents, databases and tools, with sources cited. Each agent can call only the tools you allow.
GenAIWorks Hub · Enterprise AI platform
One platform for agents, data, workflows and approvals.
Build AI agents that answer from your own documents, let anyone ask a database a question in plain English, and automate workflows with humans approving the sensitive steps. Every LLM call is redacted, policy-checked and audited.
Self-hosted with Docker Bring your own AI keys OpenAI · DeepSeek · OpenRouter
GenAIWorks Hub
Ask the agent anythingPII redacted
- 3
- LLM providers behind one governed layer
- 16
- Workflow node types
- 240
- Automated API tests
- 256-bit
- AES-GCM encryption at rest
Three workspaces, one platform
WORK, BUILD and ADMIN, each for the right people
Users get answers and approvals, builders get a full automation toolkit, and admins get control over people, keys, models and data.
WORK
Where teams get work done
- Dashboard with quick actions and recent conversations
- Agent Chat: streaming answers, attachments, email actions
- Data Explorer: plain-English questions over SQL
- Agent Jobs on demand or on a schedule
- My Approvals for workflows and agent emails
- Ask AI platform assistant
- Voice Agents, Reports and Conversations
BUILD
Where builders create automation
- Workflow Designer: drag-and-drop, 16 node types
- Workflow Monitor: runs, analytics, schedules, logs
- Automations: versioned, verified Python
- Portal Workflows: recorded browser automation
- Agent Builder with website embeds
- Custom Tools, MCP Servers, Integrations
- Data & Connections with a Data Dictionary
ADMIN
Where IT stays in control
- Users, Groups and agent / document grants
- Bring Your Own Key with per-task model overrides
- AI Security & Privacy, provider policies
- LDAP / Active Directory identity
- Tier & Usage with Feature Management
- System Prompts, System Logs, Feedback Analysis
- Append-only Audit Trail
Capabilities
Everything an AI-first team needs, already connected
Agents, data, automation and channels share the same permissions, the same audit trail and the same governed AI layer.
Data Explorer (NLQ)
Ask a database a question in plain English and get a table, chart, CSV, the SQL used, and suggested follow-ups.
Workflow Designer
A drag-and-drop canvas with variables, an AI Builder, import/export, a debug panel and a live status for each node.
Human approvals
Sensitive steps pause until the assigned person or group approves them in the Hub, Slack or WhatsApp.
AI voice agents
Phone agents run campaigns over uploaded lead lists, with retries, Do Not Call lists, recordings, transcripts and interest scores.
Executive reports
PDF, Word or PowerPoint reports where every number comes from a freshly run query. Share them with a link or schedule them.
Slack & WhatsApp
Deploy agents into Slack and WhatsApp in internal or external mode, with linked accounts and approval buttons in the chat.
Platform assistant
Ask it to create agents, workflows, jobs or schedules. It works within your permissions and asks you to confirm every change.
Portal Workflows
Recorded browser automation with agent oversight, human take-over and TOTP two-factor sign-in.
Custom tools & automations
Sandboxed Python tools and versioned automations with typed inputs and verification checks.
MCP, integrations, clouds
Connect MCP servers, an integration gallery, databases, and AWS, Azure or GCP.
Website embeds
Put an agent on any approved site with an iframe. Visitors don't need to sign in, and you can revoke an embed instantly.
Governed AI layer
Every LLM call goes through one checkpoint
Chat, Data Explorer, workflow AI nodes, the AI Builder and document learning all call models through the same provider layer, so its protections apply everywhere.
Agent Chat
Data Explorer
Workflows
Voice & Reports
Provider layer
Redact PII & secretsEnforce data policyLimits & fallbackAudit every call
OpenAI
DeepSeek
OpenRouter
Data Explorer
A 7-step pipeline for plain-English SQL
Accuracy is measured against golden question sets, and CI fails if it drops below the baseline.
- 1Link the question to relevant schema tables
- 2Retrieve examples from answers marked correct
- 3Generate SQL at temperature 0
- 4Validate: single SELECT only, plus EXPLAIN dry-run
- 5Self-correct, up to 3 attempts
- 6Execute with row limit and timeout
- 7Summarise and check every number against the rows
Workflow Designer
16 node types, from data to approval
Use {{variables}}, AI extraction that returns empty fields rather than invented values, and approvals that pause a run until someone decides.
Data Sources
DatabaseFileFolder SelectorIntegrationPortal
AI & Documents
DocumentAI ActionAI ExtractExcel Export
Flow Control
ConditionalStart LoopEnd LoopApproval
Variables & Processing
Set VariablePython ScriptAutomation
Security model
Enterprise-grade security in every layer
Secrets never appear in API responses, exports or logs. Only the last four characters are ever shown.
Encrypted at rest
Secrets, credentials, chat messages, attachments and documents are encrypted with AES-256-GCM envelope encryption. Master keys can be rotated.
PII never leaves
Personal data and secrets are swapped for reversible placeholders before every LLM call, and restricted columns never leave the system.
One RBAC model
The same role checks run in the UI and on every API endpoint. Approvals can only be decided by the assignee or their group.
Append-only audit
Database triggers keep the audit trail and the AI request audit append-only. Each AI call is logged with a payload hash.
Hardened sessions
Server-side httpOnly sessions with SameSite=Strict, CSRF protection, a login rate limit and generic sign-in error messages.
Provider policies
Each provider has a maximum data classification and limits. OpenRouter calls send data_collection: deny.
Deployment
Runs on your server, up in minutes
One Docker Compose command starts the React frontend and FastAPI backend. All data lives in a single SQLite file on your server, including the vector store for document search.
- Self-host on a VPS or on-prem, with HTTPS, firewall and backup guidance
- Bring Your Own Key: keys are encrypted and shown only as ••••last4
- Licence tiers with limits on users, agents and custom tools
- Feature flags for Portal Workflows, Integrations and MCP Servers
terminal
# configure keys $ cp .env.example .env # start the platform $ docker compose up --build ✔ schema upgraded · seed applied ✔ backend ready · frontend ready → GenAIWorks Hub is running
Use cases
Built for every team, governed in one place
Finance & accounts
Invoice extraction, approval routing, overdue analysis, scheduled board reports.
Operations
Meeting and task insights, SOP agents, portal automation, daily ops dashboards.
HR & people
Policy Q&A with citations, leave analytics, onboarding workflows.
Sales & growth
Voice campaigns over lead lists, interest scoring, follow-ups on WhatsApp.
IT & data
Governed self-service SQL, MCP tools, integrations and audit for every AI call.
Customer support
Website embeds and Slack/WhatsApp agents grounded in your knowledge base.
Need it tailored? We customise and integrate the Hub for your stack as part of our AI services.
Talk to usGenAIWorks Hub is an enterprise AI-automation platform. Teams use it to build AI agents grounded in their own documents and tools, ask databases questions in plain English, design and monitor workflows with human approvals, automate browser portals, run AI voice agents and generate executive reports, all managed by admins with users, groups, licences and AI data-security controls.
Every LLM call goes through one provider layer that supports OpenAI, DeepSeek and OpenRouter. Each task (chat, NLQ to SQL, summarisation, extraction, vision, embeddings, workflow builder) can use its own provider and model. If a provider fails or rate-limits, the next one in the fallback order is tried automatically, and the switch is logged.
Yes. Before anything leaves the system, personal data and secrets are replaced with reversible placeholders, restricted columns are never sent, per-provider data policies are enforced, and every call is audited with a payload hash. Plaintext prompt logging is off by default.
Yes. GenAIWorks Hub runs with Docker Compose on your own server or VPS. All data, including the vector store for document search, lives in a single SQLite file on your server, and you can bring your own AI keys.
It links the question to the relevant tables, retrieves examples from answers marked correct, generates SQL at temperature 0, validates it as a single SELECT, dry-runs it with EXPLAIN, self-corrects up to three times, runs it with a row limit and timeout, and checks every number in the summary against the returned rows.
See GenAIWorks Hub on your own data
Book a 30-minute walkthrough. We'll show you agents, Data Explorer and workflows on a use case from your business.