Features

What lives inside, by theme.

Grouped by the order of the work, not as a loose feature list. Each theme maps to a part of the engagement, from the first question to the month the invoice arrives.

36 features, across six themes.

The consultant

The conversation and what it produces

Led interview. It leads discovery and asks one thing at a time instead of dumping a questionnaire. You do not have to know what it needs to know.
Constructive disagreement. When the request is wrong for the scenario, it says so, with evidence and an alternative. The requirement is the business problem, never the solution you asked for.
Adapted depth. With someone who knows nothing about data, it explains by analogy from their own business. With a specialist, it goes straight to the technical point.
Memory across engagements. Facts, decisions and learnings outlive the conversation. A new engagement on the same project starts knowing what the last one found.
Distributed interview. A public link per role, so each person answers what they know. It consolidates the answers and points out where two people contradict each other.
Capture straight from the transcript. Select any part of the answer and turn it into a requirement or an open point, with the quote preserved as evidence.

The decision

Choosing on criteria, not from a catalogue

Weighted matrix. An explainable decision engine: each option is scored per criterion, with the weight in the open and the reasoning behind the score.
Exit cost per option. Reversibility, cost and time to migrate away within 18 months, and exactly what locks you in. An option that does not declare shows up as a gap.
TCO over 12 and 24 months. Cost computed by tooling, never from memory, with the source of every price in the table itself and monthly composition as a chart.
Assumption ledger. Every assumption is born with the observable condition that invalidates it and a threshold. Once broken, it points at the decisions affected.
Methodology canon. DAMA-DMBOK, Kimball, Inmon, Data Vault, Data Mesh, TOGAF and Well-Architected, curated and versioned, with the framework cited in the assessment.
Portfolio view. Opportunities that only surface across several projects at once, which single-engagement tooling cannot see.

The proof

What separates verified from written

DDL actually executed. The script runs on a disposable PostgreSQL during the audit, in a rolled-back transaction. The seal only appears if it ran without error.
Recorded red team. One reviewer attacks the decision on cost, risk and simplicity; another answers and must admit what it could not refute. All published.
Second opinion from another model. A model from a different vendor reviews the same decision. The disagreement is the product, and it is recorded rather than settled by vote.
Proof under load. A plan with synthetic data and numeric targets, run in your disposable environment. Target against measured is compared arithmetically.
Acceptance suite. One query per business requirement, with an expected result. It proves the model answers the questions, not only that it survives load.
Computed confidence seal. The fraction of available checks that were performed, with weight, score and reason per factor. A factor that does not apply leaves the average instead of entering as zero.

The context

What you already have, actually read

Multi-format ingestion. PDF, spreadsheet, Word, CSV and SQL dumps. A spreadsheet becomes text per tab, with formulas delivering the result and dates normalised.
Declared role per document. Whoever uploads says what the file is: the real as-is, a third-party proposal, a requirement. Never inferred from the content or the filename.
Tabular file profiling. A large CSV comes back as distribution per column, cardinality and extremes, computed over the whole file rather than a truncated prefix.
Search across documents. A deterministic sweep over every document in the project before opening any of them. The machine searches, the model judges.
Degraded responses declared. An empty, shrunken, duplicated file, or one that looks like SQL error output, is flagged to the consultant rather than politely summarised.
As-is from the code. Migrations, dbt and ORM from your repository, cross-checked against the database snapshot. Objects created outside review and unapplied migrations surface.

The handover

For whoever executes

Execution package. DDL, verification steps, rollback and a fenced prompt with the environment assumptions, ready for another AI or a person to run.
MCP connector. The agent on your side connects with a key scoped to a single project and closes the investigation loop without copy and paste.
Targeted investigation. It asks for the specific query that is missing, you run it, and the answer comes back bound to the question that prompted it.
Actionable exports. A data model becomes a dbt project or Flyway migrations; an architecture document becomes a CSV backlog or an infrastructure skeleton.
Document for an outside reader. The same content recalibrated for whoever was not in the room, with the reader profile declared and a reading test before it ships.
PDF and revocable link. Any deliverable becomes a PDF through the browser, or a public link you switch off whenever you want.

The governance

After the decision becomes a system

Privacy classification per column. Classification and anonymisation technique column by column, generated anonymisation views, and the classification carried into the database catalogue.
Requirement traceability. From the business requirement down to the column that serves it. A deliverable without that anchor is declared a blind spot instead of slipping through.
Drift watch. Periodic snapshots of the environment compared in code. A relevant change becomes an alert, and so does silence from the agent.
Real cost reconciled. The monthly invoice comes in and is compared against the projection. The deviation becomes calibration, and the accuracy index stays visible.
Project track record. Recommended against measured in one place, including the misses. A field with no verification shows a dash, never a zero.
Audit trail. Mutations, security events and sensitive reads recorded and queryable, visible to the organisation administrator.

Comparison

Where EVODA wins, and where it loses.

The columns compare product categories, not named vendors: the research supports the claim per category. The last four rows are where EVODA loses, and they are here on purpose.

Has it Has itPartial PartialDoes not Does notWhere EVODA loses

Capability	EVODA	Platform agentAssistants built into data warehouses and transformation tooling	Modelling toolVisual schema modellers, with AI assisting the design
Compares vendors with no commission from any of them	Has it	Does not	Does not
States the cost of leaving each option	Has it	Does not	Does not
Ties cost to the design, with 12 and 24 month TCO	Has it	Partial	Does not
Runs the DDL on a real database before handing it over	Has it	Partial	Does not
Second opinion from a different vendor model	Has it	Does not	Does not
Decision memory across engagements	Has it	Partial	Does not
Reconciles the projection against the real invoice	Has it	Does not	Does not
Document recalibrated for a non-technical reader	Has it	Does not	Does not
Applies the change in your environmentBy decision, not by limitation: whoever executes should not be whoever audits.	Does not	Has it	Partial
Tunes queries and pipelines inside the platformIf the platform is already chosen, its own agent does it better. The landing page says so.	Does not	Has it	Does not
Visual diagram editing, drag and dropEVODA generates the ERD, but it is not a canvas for drawing by hand.	Does not	Does not	Has it
Reverse engineering across dozens of source systemsHere it arrives through the snapshot your agent sends, not through a ready connector per system.	Partial	Partial	Has it

The list is long. The start is short.

Paste the schema you already have and see what it finds, before deciding whether any of these features matter for your case.

Free diagnosis