One governed core. Three engines.
Agentic search, agents, and workflows. Sovereign, Arabic-first, and accountable by design.
Everything a regulated organization needs to trust AI with real work.
One governed core: search that reasons, agents that act, and workflows that integrate, on a sovereign foundation that makes it safe.
Reasons over your knowledge
Plans a question into steps, searches across sources, checks its own evidence, and answers with exact citations.
Acts through approved tools
Autonomous and multi-agent, reading and writing in real systems, with human approval on sensitive steps.
Integrates with one API call
Add governed AI to systems you already run, and reach back into them through 100+ governed connectors.
Grounded, cited, or abstains
Every claim ties to a source you can open, or a principled “I don't know,” with a provable limit on error.
Tamper-evident audit
Every answer and action is recorded in a cryptographically linked ledger, attributed to a verified person.
On every channel
Web, Microsoft Teams with single sign-on, WhatsApp, voice, and documents, with one consistent governed answer.
Runs in your environment
On-premises or private cloud, even air-gapped. You own the knowledge, the rules, and the models.
Run by your administrators
A complete console for knowledge, tools, workflows, identity, evaluation, and branding, with no developers day to day.
Not a chatbot. Three engines, one governed core.
Search that reasons, agents that act, and workflows that integrate, on a single sovereign, governed foundation.
Agentic Search
Plans a question into steps, searches iteratively across sources, checks its own evidence, and answers from trusted knowledge with exact citations, or abstains.
- Multi-step, multi-hop retrieval with self-correction
- Exact citations, traceable to the source passage
- Abstains rather than guess, with a measurable bound
Agents
Autonomous, goal-driven agents, single and multi-agent, that complete multi-step work through approved tools, under human control.
- A manager agent that coordinates specialists
- Human approval on sensitive steps, with no double-actions
- Automatic rollback if a multi-step task fails
Agentic Workflows
Durable, governed automations, assembled visually or described in plain language, that run by schedule, event, chat, or a single API call.
- 100+ governed tools and connectors
- Triggers: schedule, system event, file, chat, API
- Durable and reversible, with every run recorded
The hard part is trust. It's engineered in, not bolted on.
Every answer and action passes one governance gate: authorized, identity-verified, validated, confirmed, and audited.
Grounding & restraint
- Answers only from approved, reviewed content
- Exact citations, traceable to the passage
- Abstains rather than guess, with a measurable bound
- Never answers from superseded or expired rules
Identity & access
- Enterprise single sign-on (OIDC / SAML)
- Acts on-behalf-of, within each user's permissions
- Step-up assurance and national-identity federation
- Need-to-know retrieval, filtered by clearance
Safe action
- Human approval on sensitive steps
- No double-actions; automatic rollback on failure
- Bounded tools and budgets per agent
- Every step recorded and reversible
Data protection & compliance
- Encryption at rest and secrets vaulting
- Egress allow-listing and PII masking, in and out
- PDPL-aligned retention and data residency
- Measured accuracy, restraint, and adversarial hardening
Reliable. Sovereign. Yours.
Deployed in your environment and run for the long haul.
On-premises or private cloud
Deploy on your own infrastructure, even fully air-gapped. No forced external egress.
Your own models
Private models run on your hardware. No foreign black box, no per-question metered bill.
Durable & reversible
Long-running automations survive restarts and resume; failed steps roll back automatically.
Observable end to end
Every AI decision is traced and inspectable, with a run inspector and single-step debug.
Enterprise identity
Single sign-on, on-behalf-of permissions, and a clearance ladder, stood up alongside it.
Business-administrator control
Your team configures agents, workflows, sources, and approvals, with no developers required.
Governed by construction. Verifiable in the product.
One governed assistant, on every channel.
The same grounded, governed behaviour on the web, in Microsoft Teams, and on WhatsApp Business, connected to the tools the work needs.
Web widget
An embeddable assistant for any portal or site, for citizens, customers, or staff.
Microsoft Teams
A native Teams assistant with single sign-on, for internal teams.
WhatsApp Business
Reach citizens and customers on the channel they use most.
Voice & documents
Speak or type, send an image or a PDF, and get an answer in kind.
MCP server tools
Connect any MCP tool or server, governed and audited like every other action.
One API call
Run any agent or workflow from an existing system through a single endpoint.
Connects to the systems already in place.
Phronexa reaches into data sources and business systems through governed connectors. Knowledge stays sovereign, and every connection is governed and audited.
Plus any REST API or MCP server. Every connection is governed and audited.
PostgreSQL
MySQL
Oracle
SQL Server
SharePoint
Google Drive
Amazon S3
Azure Blob Storage
Salesforce
HubSpot
ZendeskMicrosoft Teams
MCP serversQuestions regulated buyers ask.
Phronexa runs entirely in your environment, on-premises or your private cloud, even fully air-gapped. Your knowledge, rules, and models stay inside, and there is no forced external egress.
Answers come only from approved content, with exact citations. When the evidence isn't there, Phronexa abstains rather than guess, with a measurable bound on how often a revealed answer can be wrong.
Yes. Agents act through approved tools, within each user's own permissions. Sensitive steps pause for human approval, no action runs twice, and a failed multi-step task rolls back automatically.
Yes. Phronexa reasons natively in Arabic, including dialects, normalization, and right-to-left, and is built and tested Arabic-first rather than translated after the fact.
Through a single governed API endpoint. Your existing portals and core systems call an agent or workflow and get a structured, cited, audited result back, with no rebuild.
Your administrators do, through a complete console for knowledge, tools, workflows, identity, and evaluation. QuidexAI builds and deploys it, and can run and maintain it with you.
Sovereign, Arabic-first AI, ready to prove its value on-premises, in weeks.
Start with one high-value use case and measure it on real data.