Data & Security FAQ

The questions security, compliance, and infrastructure teams ask most often before OpsIQ goes into an environment — what it reads, where that data goes, who can see the record afterwards, and what enforces the boundaries.

A note on the architecture

It is a common assumption that Fresche hosts and controls its own AI model. That is not the architecture. OpsIQ integrates with a third-party LLM provider (Anthropic Claude), not a Fresche-owned or Fresche-trained model. For questions about the provider's model version, training practices, or security certifications, the correct source is Anthropic's own documentation, not Fresche's.

Access & Authentication

How does OpsIQ authenticate to our system? Does it use its own credentials or a dedicated user profile?

OpsIQ authenticates using IBM i user profiles. There are no separate OpsIQ credentials.

A managed install also creates an OpsIQ service profile to own and run the software, but that profile is never used to authenticate a person — see the next question. A direct install creates no profile at all and runs the services under the profile you install with. The two modes are described in Managed vs. direct install.

What IBM i authorities does OpsIQ require?

At runtime, the OpsIQ service profile is created with no special authorities (SPCAUT(*NONE)). It is also created as a non-interactive profile with limited capabilities, so it cannot be used to sign on. Its password is generated on your own system during installation, used once, and never recorded or transmitted.

It needs only ownership of the OpsIQ install directory, and *USE on the OpsIQ license library and its PW_407 data area.

Installation needs more than the running product does. The profile you install with over SSH needs *ALLOBJ, *JOBCTL, and *USE on the service profile, plus *SECADM only if you ask the installer to create the service profile for you. The installer verifies these before it changes anything, and does not use them afterwards — see Prerequisites on the IBM i. A direct install, where the services run as the profile you connect with, needs none of them.

What an individual user can see is governed by their own profile's authorities, not by the service profile's — see what prevents a query from pulling data outside a user's authorized scope below. Some tool calls require higher authority to gather their data, and the Audit Log requires *SECADM.

Does OpsIQ support SSO?

Not currently supported.

Data Access & Handling

What does OpsIQ read from our system?

OpsIQ queries IBM i system data — job logs, message queues, QSYSOPR events, and system state — in real time to generate its answers. That data is transmitted to Anthropic's API to produce the response.

What it can and cannot see sets out the boundary in full, including the application data OpsIQ has no access to.

Anthropic does not use API inputs to train its models, and deletes API inputs and outputs within 30 days by default; zero retention is available under a separate arrangement with Anthropic. Review Anthropic's privacy policy for the current terms.

Does data ever leave our environment to reach Fresche or the AI provider?

Query data is sent to the AI provider (Anthropic Claude) for processing. It is not sent to Fresche. How OpsIQ uses your system data walks through what is gathered and sent for a single question.

What prevents a query from pulling data outside a user's authorized scope?

OpsIQ uses the signed-in user's own IBM i profile permissions when querying data. The mechanism is described under output guardrails.

AI Processing

Which LLM provider and model does OpsIQ use, and who selects the model?

OpsIQ uses Anthropic Claude. Your administrator selects the model through the INVESTIGATE_MODEL setting in the OpsIQ server's configuration — a single setting that governs both the Chat Assistant and investigations. See AI model and endpoint.

Is our data used to train the AI model? What is the provider's retention and training policy?

No. Anthropic's published commitments confirm that Claude does not use API inputs to train or improve its models. Review Anthropic's privacy policy directly for retention specifics.

What output guardrails exist against inaccurate or misleading AI responses?

Three, beyond the disclaimer shown in the UI:

• The AI cannot write queries. OpsIQ ships a fixed catalogue of named, pre-written, read-only SQL statements. The AI picks a tool by name and supplies typed parameters. It never composes or sees SQL, and no endpoint accepts SQL from it. You can see the statement behind any answer for yourself — see Showing the SQL behind a tool.
• Parameters are bound, not pasted. Values are validated for type, length, and range, then passed as bound parameters to a prepared statement, so a value cannot be interpreted as SQL syntax.
• IBM i enforces authority, as always. OpsIQ obtains an IBM i profile token at sign-in and swaps to the user's profile before every query, so existing object authorities determine what they can see. A query beyond a user's authority fails on IBM i exactly as it would in a 5250 session. If the profile swap cannot be completed, the request is refused rather than run at a higher authority.

What is the impact if OpsIQ misidentifies a root cause?

OpsIQ is read-only and does not take action on the system, so a misidentified root cause affects the analysis output only, not system state. An investigation reports what it found and the evidence behind it, so its reasoning can be checked.

Verify before acting

Human verification is expected before acting on any recommendation. This is the subject of Fresche's AI Disclaimer, which you accept at install time.

Infrastructure & Encryption

Where does OpsIQ run — our infrastructure or Fresche-hosted?

Entirely on your own infrastructure. Nothing is Fresche-hosted. The tools server, which reads IBM i data, must run on the IBM i itself. The application server and web interface can run on the same IBM i or on a separate VM, whichever suits your deployment — see What gets installed for the three components, and What to install (roles) for how to split them across machines.

What encryption applies at rest and in transit?

In transit, everything is encrypted. Every OpsIQ connection uses TLS, with 1.2 enforced as the minimum protocol version. The installer generates a self-signed certificate; you can supply your own instead, as described under Your own certificate for the web UI, or replace it later — see Replacing the TLS Certificate.

At rest, OpsIQ relies on IBM i object and IFS authority rather than application-level encryption. There is one exception: IBM i profile tokens, which are password-equivalent, are encrypted with AES-256-GCM before storage. Everything else — including chat history and the audit log — is held in a local database file protected by file ownership and permissions; Where things live on the IBM i gives the paths.

What underlying components does OpsIQ rely on to access IBM i data?

OpsIQ reads IBM i data through Db2 for i SQL, using IBM's documented QSYS2 system services, over a local connection on your own system.

Are any sub-processors or third-party providers involved besides Fresche and the AI provider?

No.

Audit & Logging

What data does OpsIQ log?

OpsIQ maintains an Audit Log of user prompts and AI response summaries, along with token counts, latencies, and error messages. This log supports operator accountability and compliance review.

OpsIQ also keeps the chat history, including the skills used and the tool calls with their result sets, queries, and parameters.

Who has access to the audit log?

Only users with *SECADM authority on the IBM i LPAR can reach it through OpsIQ — see Who can see it. Any user with sufficient permissions, or *ALLOBJ, can read the underlying database file on the IFS directly.

Who has access to the chat history?

The UI shows each user only their own chat history; other users have no access to it through OpsIQ. As with the audit log, any user with sufficient permissions or *ALLOBJ can read the underlying database file on the IFS directly.

Does Fresche use audit log data for any purpose?

Fresche may ask customers to voluntarily share their audit log data to improve OpsIQ. Any such sharing is entirely optional, requires explicit customer consent, and is initiated by the customer — the product does not share it automatically.

Can we export our own audit log?

Yes. Export on the Audit Log page downloads a copy as a JSON file.

Large logs

Export captures the newest 10,000 entries. If your log holds more than that, OpsIQ tells you so after the export. See Exporting.

Does OpsIQ monitor AI output quality, latency, or error rates over time?

Not currently supported.

Retention & Deletion

How long are audit logs and chat history retained?

OpsIQ does not currently expire or delete audit entries automatically. The log is append-only and retained for the life of the installation.

How can we delete audit and chat history data?

Audit data cannot currently be deleted through the UI.

Chat history can: each chat is deleted by its own user from the Sessions page.

If you are removing OpsIQ's data for compliance reasons, note that uninstalling removes the install directory — chat history and logs included — but deliberately leaves pre-upgrade snapshots behind, and those contain copies of the same data. What gets removed covers what to clear by hand.