Three steps, and the source named every time.
DocuDriver reads what you upload, proposes the data dictionary from it, and answers questions from the approved version. Nothing is invented, nothing is published without approval, and every answer says where it came from.
Upload what you have: policies, procedures, models, reports, SQL, packages, and the documents around them. The product reads the structure, proposes the dictionary, and answers from it with the source named every time.
Upload the deliverables
Power BI files and models, SQL, SSRS reports, SSIS packages, dbt projects, notebooks, Office documents, PDFs, and code. Extraction keeps structure and metadata only: table names, fields, measures, queries, lineage. Never a data row.
Approve the dictionary
Every table, field, measure, and query is proposed as a draft with its definition and source. Your people approve, edit, or reject. The product never invents a definition and never publishes one nobody approved.
Ask, and get a cited answer
The assistant retrieves only from documents the member is allowed to ask, cites what it used, and says so when the answer is not in the documentation. Access rules are enforced at the database, not in the prompt.
with movers as step).Retention you can defend
Review, archive, and version rules per collection, legal holds that freeze everything under them, and a purge window with a restore queue. Nothing is ever silently deleted.
A dictionary that proposes, and people approve
Definitions come from the delivered files themselves and carry their source. One dictionary per collection, retired when it empties, exportable for anyone who needs it on paper.
An assistant that cites, or says so
Answers come first from documents the member may ask, with the source named every time. When the documentation does not say, the assistant says so plainly before offering what it knows in general.
Examples in action
Three situations DocuDriver has been built around, written out so you can picture a question and its answer. Every organization runs in a workspace of its own, with its own documents, members, and rules, and gets the same product and the same assistant, whether it is a college, an agency, a support desk, a hospital group, a city department, or a law practice. If your documentation is large and exacting, there is an example here that looks like your week. Open one to watch the assistant answer; the exchanges are scripted from fictional documents, in the product's own interface.
A college analytics teamReports, SQL, models, and the definitions behind every number, answerable for the people who build and the people who ask.
An institutional research office carries years of reports, SQL, models, and definitions, spread across drives, tickets, and people's heads. In one workspace the documentation has a home, the data dictionary is proposed from it, and questions are answered from the approved version.
What matters here
Reports that outlive the person who built them
When an analyst leaves, the reports keep running but nobody can say what a field means or why a filter is there. The report, the model, and the SQL are read, and the dictionary is proposed from them with the source of every definition.
Definitions people can agree on
Enrollment, retention, credit hours, and full-time status mean different things in different offices. One approved definition per term, with its history, so the registrar, institutional research, and finance point at the same sentence.
Compliance reporting without the scramble
IPEDS, state reporting, and accreditation reviews ask the same questions every cycle. With the queries and their definitions in one place, "how did we calculate this last year" is a question, not a search.
What the workspace does in this situation
- Reads Power BI files and models, SQL, SSRS reports, SSIS packages, dbt projects, notebooks, and Office documents, keeping structure and metadata only. Never a student record.
- Proposes the data dictionary from those files for approval, so nothing is published that nobody checked.
- Groups departments inside one workspace with roles and a hierarchy, so institutional research, the registrar's office, and a dean each see what they are granted.
- Signs people in through the institution's own identity provider and provisions them from the directory on the Institution plus SSO plan.
Fewer people asking the same person, faster onboarding for new analysts, and a dictionary you can hand to an auditor.
An insurance agencyPolicies and their changes, who each one applies to, and the clause in hand when a client asks.
An agency lives inside documents that change under it: carrier policies, underwriting guidelines, state rules, internal procedures. The workspace keeps the current version and the history, shows what changed, and answers questions about how a policy applies, with the clause in hand.
What matters here
Knowing which version is in force
A guideline updated in March should not be answered from the February copy. Document attributes record effective dates, status, and what supersedes what, and the assistant answers from the version that applies.
Seeing what changed
When a carrier or a regulator issues a new version, a review compares it against what you had and lists the differences, section by section, as a findings report the team can work through.
Tying policy to population
Which customers, products, or states does a rule touch, and where does that show up in the systems that hold them? The dictionary describes the fields and tables in your databases, so the assistant can connect a rule to the fields that would identify who it applies to.
What the workspace does in this situation
- Records effective dates, status, supersession, and "applies to" on each document, and answers from the version in force.
- Compares two documents in the chat, or runs a review of a new version against the current one and produces a findings list and report.
- Watches external sources you register (a regulator's page, a carrier's bulletin) on a schedule you choose and tells you when they change.
- Marks one collection as your own and another as a reference, so comparisons know which side is yours.
Agents answer from the current version, changes are found before they become errors, and the record of who knew what and when is already written.
A sales and support deskThe right procedure in the moment, from the approved version, with the step named.
A help desk and a sales team work from policies and procedures that are long, strict, and changing. An assistant beside each person answers from the approved documents, names the step, and says when the answer is not written down yet.
What matters here
Consistency under pressure
Two agents should give the same answer to the same question. When both ask, they get the same procedure, from the same approved version, with the section named.
New hires who are useful in days
A new agent does not need to know where everything is. They ask, they get the answer and the source, and they learn the documentation by using it.
Knowing what is missing
Every question the documentation could not answer is a gap. Those show up for the people who own the procedures, so the next version covers them.
What the workspace does in this situation
- Holds procedures, policies, scripts, product documentation, and escalation rules in one workspace, with roles that decide who sees what.
- Marks plainly when an answer comes from general knowledge because the documentation does not cover it, so a gap is visible rather than guessed over.
- Keeps the approved version current, with review dates and versions, so an outdated procedure is retired rather than lingering.
- Logs every question and answer in the audit trail, which supervisors can search and export.
Faster first answers, fewer escalations caused by guesswork, and a documentation set that improves from the questions it could not answer.
Examples in depth
An answer is a starting point, not the end. Everything the assistant names is a link to its definition, and every claim carries a footnote to the passage it was read from. Open one to watch where a click goes.
From a name in an answer to its definitionEvery table, field, measure, and query the assistant names is a link to its entry in the data dictionary.
When an answer names something the dictionary knows, the name is a link. One click opens the entry in a new tab: the approved definition, where it came from, and for a table, every column with its type and definition. The conversation stays where it was.
Ready Course Pct is calculated in CourseReadiness.pbix over the view rpt.course_readiness, which the SQL builds from the course and section tables.1
The view carries one row per course and term with its readiness status and whether it is active; the measure divides the Ready rows by the Active or Ready rows for the selected term.2
What to notice
- Names are linked only when the dictionary holds an approved or proposed entry for them, so a link is a promise that a definition exists.
- The entry page shows the definition, the formula where there is one, the columns of a table, and what uses it and what it uses.
- Open the source at this passage takes the reader from the entry to the exact place in the file the definition was read from.
From a footnote to the passage it came fromThe small numbers in an answer are the passages retrieval read; each one opens the source at that passage.
Every claim in an answer carries a footnote number. Hover it and the passage is named; click it and the source opens in a new tab at that passage, highlighted, with the passages around it for context. Nothing in an answer is unsourced.
Ready Course Pct is calculated in CourseReadiness.pbix over the view rpt.course_readiness, which the SQL builds from the course and section tables.1
The view carries one row per course and term with its readiness status and whether it is active; the measure divides the Ready rows by the Active or Ready rows for the selected term.2
What to notice
- A footnote points at one passage of one document, the same text retrieval searched, so what the reader sees is exactly what the assistant read.
- The passage view shows the passage number, where it sits in the file, and names of other sources and dictionary entries inside it as links of their own.
- The From chips under an answer open the whole document on the Sources page; the footnotes open the exact passage.
Beyond the dictionary
Once the documents are in, the same workspace does the work around them.
Versions and effective dates
Each document can carry an effective date, a status, what it supersedes, and who it applies to. The assistant answers from the version in force, and the old one stays on the record.
Comparisons and reviews
Compare two documents in the chat, or run a review of a new version against the current one and get a findings list and a report your team can work through.
Watched external sources
Register a regulator's page or a vendor's bulletin and the product checks it on the schedule you choose, telling you when it changes.
Retention and legal hold
Review, archive, and version rules per collection, a purge window with a restore queue, and holds that freeze everything under them.
Departments as groups
Roles, collection grants, and a hierarchy decide who can see, ask, or download each source. One workspace, every department inside it, each seeing its own.
Your own model, if you want it
Use the plan's included questions, or bring your own AI model key for unlimited questions with nothing passing through our model account.
See it on your own documentation
A demo takes about 40 minutes on our example workspaces, or on a policy, a report, or a folder of SQL you bring along. Anything you bring is removed when the demo ends.