DovaAI: contextual help and Mapping proposals

Purpose and scope

DovaAI uses the authorized tenant, current module and view to explain Dova controls, components and configuration. It can guide you through the 36-component Mapping catalog and propose a reviewable design. Interface context does not grant access to other tenants, hidden data, credentials or operations outside your permissions.

Responses follow the message language: Spanish, English or Portuguese. You can explicitly request “in English” or “em português”. When a message does not clearly identify another language, the available language reference is retained. You do not need to select an assistance mode or language first.

Before you start

Confirm tenant and permissions. Schema attachment requires Mapping access; draft creation requires its corresponding authorization. Service configuration belongs to the authorized administrator and stays outside the conversation. Never paste keys, passwords, private certificates or customer information.

Usage allowances depend on the effective plan and contract. Starting another conversation does not reset consumption or grant more capacity. A configured service can still reject requests due to availability or limits.

Request and review a Mapping

1. Describe the goal, fields and expected result. For example: “Keep name and amount; change only the exact name value jorge to Jorge and prepare CSV output.”
2. If file context is needed, choose Attach data file and select a synthetic CSV, TSV, JSON, JSONL or XLSX sample. Inspect inferred names and types; inference does not replace your business schema.
3. Review components, connection directions, settings and missing fields. For exact replacement, Replace values should use field=name, find=jorge, replacement=Jorge, matchMode=exact and caseSensitive=true. This does not capitalize arbitrary names.
4. Confirm creation only when the proposal matches the goal and authorized location. A proposal is not an executed Mapping. If multiple drafts are offered, review each artifact and its required permissions.
5. Open the created draft and manually complete execution files, Connections, paths and parameters. Inspect missing settings; attaching a chat sample does not place a file on the worker.
6. Use the normal Validate, Save draft and Publish lifecycle. Then perform an authorized synthetic run and accept its output in Monitor. Assistance does not automatically publish, schedule or execute the proposal.

What an attachment provides

The service temporarily inspects the sample to derive its name, format, field names, inferred types and source settings. The AI context contains that structure, without sample rows or original file bytes. File, column and sheet names can also be sensitive: use laboratory data.

The schema is not saved as a run input. Supply the real execution file through Run or the appropriate bindings. A chat attachment does not establish connectivity, destination authorization or worker compatibility.

Verify the result

For the example, inspect Flat File input → Field selector → Replace values → Flat File output. Field selector retains only name and amount in records and output schema; Replace values changes the specified exact value. Check that other capitalization and different names remain unchanged, then compare the actual CSV after execution.

You can move, minimize or open chat in another window where supported. Context adapts when switching modules within the same authorized tenant. Identity, tenant or permission changes invalidate stale actions; continuity after sign-out or browser restart is not guaranteed.

Limits and recovery

Samples must be nonempty and at most 15 MiB; JSON and JSONL are limited to 1 MiB. The schema requires 1–32 unique field names. CSV/TSV requires a valid header and UTF-8 encoding. Attached structure expires after 15 minutes and can be lost when the service restarts; attach it again when expired.

An oversized schema or invalid context can be rejected. Correct the sample rather than increasing permissions. For exhausted allowance, inspect the plan. For an incomplete proposal, clarify the goal and fill missing fields in the editor. If creation outcome is uncertain, find the draft in Repository before repeating confirmation.

Continue with a controlled test

Use name,amount,internal_note with two synthetic rows whose names are jorge and Jorge. Expect only name and amount in output; both names become Jorge, amounts remain unchanged, and internal_note is absent. Continue with Mappings and Monitor. For unresolved issues, use the Cloudata support form without secrets or customer data.