AI agents your customers can trust, on a platform you brand.
Agentic Operations is delivered as source code. An agent works for a customer inside their own workspace, a person stays in control, every step is on the record, and the platform carries the acquirer's brand.
Agentic Operations is Primanza LLC's own product, filing number 2026-002018263: built and tested by the company on its own account. It is delivered as source code and is not operated by Primanza. It is not client work.
At a glance
- Status
- Version 0.2.0, released as source code on October 19, 2026
- Delivery
- Source code, by acquisition or license.
- Runs on
- The acquirer's own infrastructure, on Docker
- Pricing
- No price is published. Terms are discussed directly with the company.
- Contact
- sales@primanza.com
- Built by
- Primanza LLC
- Registry
- PRZ-002
The demo film
Example data
Why Agentic Operations exists
A company that wants to sell AI agents needs the same layer underneath before it can build any one of them: tenants and permissions, a person who approves what an agent should not decide alone, a record of everything an agent did, connections to a customer's systems, and a choice of model providers. Agentic Operations is that layer, delivered as source code. The acquirer's brand, prices and accounts sit on top of it and are never hardcoded into it.
Software companies that want to build and sell an AI product family of their own, and the integrators and resellers who deploy it for customers.
How it works
The demo film follows these five parts, in order.
The agent
A customer's administrator signs in to their own workspace and tests an agent from the console. In the film the sales agent answers a stock question from the demo ERP and a warranty question from a manual it cites.
Control
A quotation above the discount limit stops and waits for a person. The decision page shows why it stopped; the person comments and approves, and the comment stays in the audit history.
People
When the agent cannot solve a fault code it hands the conversation over. The person gets the customer's message, what the agent tried and why it escalated, takes it, and replies.
Proof
Every run records its model calls, tool calls, knowledge lookups and tokens. The audit trail records who changed what, and "Verify the chain" shows whether a record was edited or removed.
Your platform
An operator changes the product name and colours in one save, with no code change, and the console and the sign-in page follow. The same console manages customers, workflows, knowledge, connectors, analytics and usage against plan limits.
What a company builds on it
A sales agent
Answers from ERP stock and prices and from a warranty manual, and writes quotations; one above the discount limit waits for a person.
A technical-service agent
Works a service queue and hands the conversation to a person when it cannot solve a fault.
Durable workflows
Visual workflows with triggers, agents, tools, approvals, human tasks and waits, versioned and released into each environment.
A knowledge assistant
Answers from the company's own documents and cites the document it used.
A product family
The platform's scope names sales, customer-support, technical-service, back-office, web and messaging agents, workflow automation, enterprise knowledge assistants and vertical products.
What a buyer asks first
The questions a software company asks before it acquires or licenses a platform, answered from the release's own records. Those records are kept in a private repository and a buyer receives them with the code. Where a summary here and a record differ, the record governs.
- What is finished?
- Release 0.2.0 of October 19, 2026: 92 of its 95 capabilities are verified in the project's own tests, 1 is partial and 2 are foundations. The records say which three and why.Source: Pre-sale readiness record, section 3
- Has it run in production?
- No. It has never carried a customer's traffic. Its cloud references were never applied to a real account, and no hosted model vendor was called in its own runs, which used a deterministic mock, Ollama and in-repo fakes.Source: Pre-sale readiness record, assessment and section 12; known limitations, deployment and agents
- Has anyone outside the project tested or certified it?
- No independent third-party penetration test has been performed, legal review has not happened, and it has not been audited or certified against SOC 2, ISO 27001, HIPAA, PCI DSS or any other standard. Its own scans, reviews and live-stack assessment are documented.Source: Pre-sale readiness record, section 6; known limitations, introduction and identity and security
- How is one customer's data kept from another's?
- Every tenant row is guarded by row-level security in PostgreSQL, and the tenancy test suite covers every table. A dedicated tenant means a separate installation.Source: README, architecture; known limitations, deployment and operations
- What does it run on?
- One NestJS API and one worker over PostgreSQL 18 with pgvector and Valkey, with a Next.js console, on the acquirer's own Linux and Docker infrastructure and with no paid service in its core. A Helm chart and Terraform references for AWS and Azure are included; the Terraform stacks were checked with mocked providers and never applied.Source: README, architecture; known limitations, deployment and operations
- Which AI models can it use?
- Model providers sit behind ports, with adapters for OpenAI, Anthropic, Gemini and Ollama, and a deterministic mock as the default. The live tests against hosted providers need the acquirer's own keys and were not run by the project.Source: README, quick start; pre-sale readiness record, section 4
- Who owns the brand and the customers?
- The acquirer: its branding, customer relationships, pricing, hosting, cloud and model accounts, integrations and extensions. The platform never hardcodes them.Source: README, product
Checked against those documents on October 23, 2026.
What a buyer reads first
The documents in the repository for a buyer, in the order its README gives. They are delivered with the code.
Buyer onboarding
Technical due diligence
Source code handoff
Pre-sale readiness
Known limitations
Final acceptance
Talk to us
Questions about source code, an acquisition or a license go to the sales desk.
- Contact
- sales@primanza.com
- Built by Primanza
- Agentic Operations is Primanza LLC's own product, filing number 2026-002018263: built and tested by the company on its own account. It is delivered as source code and is not operated by Primanza. It is not client work.All products · About Primanza
Discuss your project.
Describe the problem, the systems involved and what a good outcome looks like. The discovery conversation starts from there, and the proposal puts the scope in writing.
You write
Through the form or to hello@primanza.com: the problem, the systems involved, what a good outcome looks like, and any constraints you already know.
We talk
A discovery conversation about the workflow as it runs today and the outcome you need. A reply comes within two business days, Monday to Friday.
The proposal
Deliverables, assumptions, dependencies, acceptance criteria, timeline and commercial terms, in writing, for you to accept or change.