Docyrus DSQL Query Design

DSQL (Docyrus Structured Query Language) is a read-only, PostgreSQL-compatible SQL dialect that queries logical data sources — not physical tables. Tables are named appSlug.dataSourceSlug. This skill covers the full workflow: schema discovery → query authoring → execution.

For the complete language reference (all supported functions, tenant pseudo-functions, row limits, rejection rules), see references/dsql-language-reference.md. For common query patterns and examples, see references/query-patterns.md.

Workflow

Follow in order.

1. Confirm auth.

   docyrus auth who --json   # confirms session + tenant name
   

   No session → docyrus auth login first.

2. Identify the data sources. If app/source slugs are already known, skip to step 3. Otherwise:

   docyrus apps list --json                          # list apps → grab appSlug
   docyrus dsql schema app <appSlug> --json          # all queryable sources in the app
   

   The schema response includes a compact CREATE TABLE appSlug.dataSourceSlug (...) DDL per source — this lists all queryable field slugs and -- references join hints.

3. Fetch the schema for the exact sources you'll query. Three options — pick the smallest:

   Situation	Command
   Know appSlug + dataSourceSlug	docyrus dsql schema data-source <appSlug> <dsSlug>
   Know one or more data source IDs	docyrus dsql schema data-sources --ids <id1,id2>
   Need all sources in an app	docyrus dsql schema app <appSlug>

   Always read the schema before writing joins — field slugs and reference hints are only visible there.

4. Write the DSQL query. Follow the rules in references/dsql-language-reference.md. Key constraints:

   • Tables are appSlug.dataSourceSlug (e.g. base.task, crm.contact)
   • Always alias every table; qualify all columns in multi-table queries
   • Bare * only when exactly one source is in scope
   • Default limit 100 applies if omitted; max 1000

5. Run it.

   docyrus dsql query "select t.id, t.subject from base.task t limit 10"
   # or from a file
   docyrus dsql query --from-file ./my-query.sql
   

6. Quick path — AI generate + run in one shot. For natural-language questions where you don't need to handwrite SQL:

   docyrus dsql ask "how many open tasks per project this month?"
   

   ask calls the DSQL Generator agent (discover schema → write DSQL → execute → return results). Use generate if you only want the SQL without running it:

   docyrus dsql generate "list contacts created in the last 30 days with their owner names"
   

7. Explain the result. Summarize what the data shows; note any relevant limit truncation or empty-result reasons.

Command cheat-sheet

Schema discovery

# All sources in an app
docyrus dsql schema app crm --json

# Single source
docyrus dsql schema data-source crm contact --json

# Multiple sources by ID
docyrus dsql schema data-sources --ids "019c48d0-...,019c48e0-..." --json

Execute a query

# Inline SQL
docyrus dsql query "select p.id, p.name, p.status from base.project p order by p.created_on desc limit 25"

# From file
docyrus dsql query --from-file ./report.sql

# JSON output
docyrus dsql query "select count(*) as n from base.task t" --format json

AI-assisted generation

# Generate only (returns { prompt, query })
docyrus dsql generate "total time logged per project last week"

# Generate + run (returns { prompt, query, data, meta })
docyrus dsql ask "how many tasks are overdue by assignee?"

# Override the DSQL generator agent
docyrus dsql ask "..." --agentId <custom-agent-id>

Critical rules

• Table notation is appSlug.dataSourceSlug, never a physical Postgres table. Bare table names (task, contact) are rejected unless they are CTE aliases.
• Always alias every table (from base.task t). Qualify every column in joins — DSQL rejects ambiguous unqualified column references.
• Bare * is only valid when exactly one logical source is in scope. select * from base.task t join base.project p ... is rejected — use t.* or list columns.
• Read-only only. INSERT, UPDATE, DELETE, CREATE, DROP, and any DDL are rejected outright.
• No physical tables, no schema-qualified functions. select * from public.tenant_user or select pg_catalog.now() — both rejected. Use tenant.user for users and the approved function list.
• Tenant pseudo-functions (tenant.current_user_id(), tenant.user_in_current_teams(userId), etc.) are the only way to filter by current session identity securely. Don't substitute hard-coded UUIDs for them.
• tenant.user and tenant.enum are system sources for resolving user names/emails and enum labels/colors. Join on u.id and e.id respectively.
• Inherited base fields: base.task and base.event expose base.activity fields (e.g. record_owner, created_on) — query them directly on the child source.
• Row limit: default 100 if no LIMIT clause; hard cap 1000. A user-written smaller limit is preserved. LIMIT ALL does not bypass the cap.
• Schema first: never guess field slugs — fetch the schema and read the DDL. Wrong field slugs return a DSQL error.
• docyrus dsql ask is the one-liner path: natural language → AI-generate → run → results. Use it when handwriting SQL is unnecessary.
• CLI-only — do not use in frontend pages. The underlying endpoint is not documented in Swagger and is not intended for use in implemented React/frontend application code. All DSQL queries must go through the docyrus dsql CLI commands.

References

• references/dsql-language-reference.md — Full DSQL language spec: supported clauses, functions, tenant pseudo-functions, table resolution rules, rejection list, row limits.
• references/query-patterns.md — Common query patterns: aggregates, joins, CTEs, tenant filters, enum lookups, date ranges, and error reference.