nondro

Create a draft link type between two object types.

create_link_type

Create a draft object type — the nouns of the organization.

create_object_type

Column names and types for ONE table/view of a source. Use this before writing queries — never guess column names.

describe_dataset

What a given artifact kind's spec must contain: required and optional fields, what each means, and a worked example. Call this BEFORE authoring or editing an app / action_type / pipeline / playbook — never guess the shape of a spec.

describe_artifact_kind

Contribute a proposal, question, correction or challenge without rewriting consensus.

discuss_work

Exercise a pipeline's NEWEST version (draft or live) against real sources without writing anything: reads and transforms run on a small sample, loads and pushdowns are refused. Returns what each step would do. Needs invoke.

dry_run_pipeline

Replace a supported complete draft, conditional on its latest version.

edit_draft

Fetch one result of search by its id: its definition, lineage and, for a pipeline, its recent runs, as text.

fetch

Fill a Word template with data and get the document, as DOCX or as PDF. The template is a .docx with fields ({{ name }}, {% for r in rows %}…{% endfor %}, {%tr %} for table rows) that the organization already has in a document source (template_source + template_name) or in an app's files (template_app + template_path). Values are escaped and the template runs in a sandbox. The result is saved in YOUR document folder and returned with its SHA-256, so what circulates can be proven to be what was made.

render_document

Create an explicit branch of work; this does not branch or copy catalog artifacts.

fork_work

Mechanically gathered facts about the integrated data, for judging what the organization's nouns might be: per dataset, whether a key column is unique and never empty (checked with counts), foreign keys declared or sharing a key's name, the values of status columns (only columns not marked sensitive, only when there are few), what pipelines read and produce, and a guess at each table's shape (lookup, joining table, event, thing). It reads names, keys, counts and those status values; it sees no meaning, and its guesses are about tables, not about the organization. It runs in the background; read_ontology_proposal returns it.

gather_ontology_evidence

Ask for a connector: the way to reach data on an organization's own server (a database, a folder) without moving it. Returns ONE line for a person to run once on that server; it installs the connector, which creates its own database login for exactly the database and schemas named here, and trades a one-time code for its own key (no key ever passes through anyone). Once it runs, what it shares becomes sources by itself and Sonda proposes what is in them. Give the line to the person; the code in it works once, for 24 hours.

add_connector

Read one object of the index by its obj_fqid, under the same lens.

get_object

Recover authorized understanding for a task, with evidence and continuation. Resolve an optional project name without guessing. Read further pages when coverage is incomplete. Context is assembled; it creates no memory or work.

get_task_context

Start here. The organization's context exactly as nondro's embedded copilot receives it (`pack`, with its `pack_hash`): the rules it works under, the catalog, current knowledge, dataset profiles, this connection's conversation, past episodes and the playbooks to follow. Also your identity, permissions revision and a page of permitted definitions and knowledge. Pass `query` (what the task is about) to focus the knowledge, and `project` to put that project's work first in the pack and keep the catalog page to it. `budget_bytes` bounds everything except `pack`, which always comes back whole.

get_context

Get the canonical ranked knowledge pack used by every nondro interface.

get_work_context

Hand a work thread over: you stop being the one who writes to it, and whoever resumes it next (resume_work) takes it over. Include `handoff_note`: a short summary you write of the work since your last note (what the person asked for, what was decided and done, what is still open). The next participant sees it with the work's checkpoint. Changes only who holds the thread in nondro.

handoff_work

Ask the index to read one object type now: its live mappings, then the live links that touch it, on the worker, without waiting for the next cycle. Follow with read_index_status.

sync_index

List the datasets (tables/views) a source exposes.

introspect_source

List the actions your organization has defined in nondro for its objects (for example, approving a request), with the kind of object each applies to. Nothing is run or proposed.

list_actions

The files of an app's or a script's version (its newest draft unless a version is given).

list_app_files

List this user's conversations on the platform (titles, scope, when), including the work an AI did for them through a connection (marked via: connection). Your memory spans them — use read_conversation to recall one.

list_conversations

List registered data sources and the connection refs available for new ones.

list_sources

List the documents in a files source (name, kind, size, modified).

list_documents

Discover registered document names before requesting evidence excerpts.

list_source_documents

List the work threads you may see (pieces of work people and AIs carry out in nondro), with their titles, subjects, state and who holds them, filtered by words in the title or by the subjects or project they concern. For finding work to read or to continue.

list_work

Map a dataset's columns onto an object type's properties.

create_mapping

Move a definition into a project, or out of every project so the whole organization shares it. Grants can name a project, so this changes who reaches it: you need author permission where it is and where it goes.

move_to_project

Change PART of an artifact's spec, without restating the rest, as a NEW DRAFT VERSION. Use this instead of update_artifact_spec whenever the spec is large and the change is small: a LEFT JOIN added to one transform's SQL, two projections renamed, a surface's query fixed, a schedule changed, a step appended. update_artifact_spec makes you re-emit every character of the spec, so on a 254-column pipeline a three-line change costs 38,000 characters of SQL retyped from memory and one lapse silently deletes a column. Keep update_artifact_spec for small or empty stubs and for restructuring a spec wholesale. Read the artifact first and COPY the text you are changing; never type it from memory. Every edit must land exactly: a `find` that matches zero times, or a different number of times than expected_count, refuses the WHOLE patch with the counts and writes nothing at all. The result is a draft like any other — a human still promotes it through the gate.

patch_artifact_spec

The address where a draft version of an app can be opened before it is live: only who may work on the app, and the people its audience names, can open it. Also says where the live version answers.

preview_app

Profile a registered dataset: columns, row estimate, low-cardinality values.

profile_dataset

Put a version of a definition LIVE, as you, through the same gate as nondro's Promote button: the rulebook runs on that version and refuses it when anything blocks, and you need promote permission on the definition. An earlier version rolls a change back. Promoting a pipeline registers the tables it writes as datasets, and it runs on its schedule from then on. Use it when the work you were asked to do needs this live; use request_promotion when a person should decide instead.

promote_definition

Propose that one of your organization's actions be carried out on one object. Nothing runs: nondro files a proposal, and a person with the right to decide approves or rejects it in nondro; only an approved proposal runs, as the action defines. The proposal is checked against the action's current version and its declared parameters.

propose_action

Read instances of an object type from the object index — the GOVERNED path. Typed objects already narrowed to the rows and properties you may see, with a note naming what your scope withheld. Reach for this first: it is the ontology, so the answers are the organization's own nouns rather than whatever a table happens to be called. run_readonly_query is the raw path.

query_objects

Read the transcript of one of this user's PAST conversations (their own only — never anyone else's). Text messages only; tool chatter is elided.

read_conversation

Read a permitted definition. Returned definitions can be partial.

read_definition

List a definition's versions — who wrote each, when, why, which is live — or, with version, read that version's spec.

read_versions

Read a document's text, a few pages at a time. Long documents are paged: read what you need, then read further pages. Never assume content you have not read.

read_document

Read a pipeline's recent runs: when, who or what started them, how many rows each step moved, and why a failed run failed. Newest first. Runs not finished yet come in `in_progress`: a waiting one with how many long jobs are ahead of it, a running one with the step it is at and since when.

read_runs

Read a catalog or indexed object subject under the current row and field lens.

read_subject