For IT and AI managers
How results are checked and how AI use is recorded.Where the data stays and who can see what.
Evaluation and acceptance
Afunana checks named objects, file definitions and relationships against source and metadata.These checks cover the overview, processes, system areas, glossary, files and relationship views.Build records report completeness and failures; each verification record states its scope and limits.
Build checks compare document references and claims with source and metadata.Reports distinguish findings from missing or unsupported evidence.On request, an AI assessment reads the source, then compares the document to it.It checks supporting evidence, contradictions and important omissions.The assessment retains the versions examined and its run details.
A reviewer records OK, a remark or a problem for the version examined.The record keeps the findings as they were at review time.Later changes never rewrite that evidence.An analyst can sign off a program or file document; a later change to it is flagged.A passage report opens a version-linked case for investigation and review.Corrections are verified; approved notes never silently overwrite the original.
Chat checks examine citations, named objects, incomplete evidence and truncated answers.For streamed answers, the checks follow the displayed text.Each problem found is listed in the Verification window with its evidence.
AI governance
Governance records how AI is used, who is responsible and what needs review.Every model call is recorded with its user, collection and tokens.
Record the deployment's intended use, scope and named responsibilities.Keep risk and impact assessments, review dates and open items.
Record AI incidents and concerns with status, resolution and who closed them.Documentation issues stay linked to their passage-level review cases.
Audit records name who performed each security and admin action.Usage reports, governance exports and SIEM forwarding support internal reviews.Prompts are versioned files, and most new capabilities ship switched off.
Afunana keeps the records and oversight your compliance work needs.Obligations such as the EU AI Act depend on use and deployment.NIST AI RMF and ISO standards are separate references.
Change planning and compiling
Afunana finds the affected programs and plans the work, linked to source.Proposed edits are applied to copies in the developer's library, and the source differences are shown.Compile-check results are recorded with the plan; a missing compile is not a pass.A failed compile gets up to two recorded fix rounds.Each plan carries its regression tests; a risk label can be switched on.The developer executes the plan after reading that evidence.
With personal sign-ons, IBM i saves go to each developer's mapped library.They run under that developer's own profile.Until sign-ons are set up, saves go to the collection's source library.Every member save is kept in the collection's own history.
An unresolved finding or user report stays open in its review case.Reviewers record the decision and the fix against the same case.
Security and data handling
Afunana runs on your infrastructure with Docker or Podman.Source, documentation and the search index stay in that installation.Customer-managed databases and reverse proxies are supported.
Relevant source and context go to your configured AI service or gateway.A model hosted inside your network keeps that processing inside.An external AI assistant sends what it retrieves to its own provider.
Documentation uses source, definitions and supporting documents.It does not require production business records.The extraction reads source and catalogues, never business data rows.Table browsing is a separate read-only screen for users entitled to the collection.
One script installs and updates Afunana to an exact version you choose.It runs on RHEL, AlmaLinux, Rocky Linux, Ubuntu or Debian.If the AI provider is down, existing documentation still works.New AI answers resume when the provider is back.
Identity and permissions
Afunana supports local accounts, SAML and OpenID Connect.Administrative access is separate from ordinary use.Sessions expire, can be revoked and require sign-in after a restart.Local accounts can require an authenticator code.
Entitlements decide which collections a user can read, on every surface.Six built-in roles decide what each person may do.AI assistants sign in to the MCP server with OAuth.MCP requests check the caller's access to a collection before answering.
Cost and benefit
Agree representative tasks, each with an acceptance standard.Compare effort by role: preparation, source reading, expert review, correction.Use tasks of similar complexity.Repeating a known task overstates the gain.
Include licensing, infrastructure, preparation, training, AI usage and admin.Usage records and rate estimates cover the AI share.Invoices and staff costs complete it.
Time freed for other work is additional capacity.It becomes direct budget savings when it reduces expenditure.Report quality gains and reduced dependence on key people separately.Expand beyond the first collection when quality, effort and cost justify it.
Deployment assessment
We agree the boundary, data handling and acceptance criteria with your people.
