Skip to main content

Documentation Index

Fetch the complete documentation index at: https://datost.com/docs/llms.txt

Use this file to discover all available pages before exploring further.

A Context is a persistent conversation with Datost about your data. It remembers every message, every query it ran, and every artifact it produced, so you can iterate across many turns without re-explaining yourself. Contexts are the default landing page of the Datost web app at /context and are the same conversation objects that power Slack threads — a question you ask in #analytics lives in the exact same history as one you ask in the browser.

​
What a Context includes

Each Context is a long-lived conversation with:
  • Messages — your questions and Datost’s replies, in order.
  • Runs — each assistant reply is backed by an aiRun, which records every tool call, step of reasoning, and intermediate result the agent produced.
  • Artifacts — charts, tables, CSVs, and generated notebooks the agent wrote during a run. These are persisted and downloadable long after the run completes.
  • Source selection — which warehouses, repos, log providers, docs, and Slack channels are in scope. RBAC is enforced on the server; you can’t widen your own access by toggling sources.
Contexts in the web app and threads in Slack share the same backing data model. A Slack thread about Q3 churn and a browser Context about Q3 churn are both conversations rows — they just render differently.

​
Starting a Context

1

Open the web app

Navigate to the Datost web app. /context is the default route, so a fresh Context is waiting for you.
2

Ask a question

Type a question into the composer — “show me the largest tables by row count”, “what errors spiked in the last hour”, or anything more specific. Hit Enter to send, Shift+Enter for a newline.
3

Watch it work

Datost streams its reasoning live: which tool it’s calling, what SQL it’s running, what it found, and what it decided to do next.
4

Follow up

Ask the next question in the same Context. The agent carries the full history forward, including earlier query results.
You can also kick off a Context from Slack by @-mentioning @Datost in any channel or DM. The resulting thread is a Context you can later open in the web app to review artifacts, re-run steps, or branch the conversation.

​
Multi-step reasoning and tool use

Datost is an agent, not a single-shot text model. Inside one reply it may:
  • Run SQL against your warehouse, read the results, and follow up with a second query.
  • Search your repos for a dbt model to understand how a metric is defined.
  • Scan log connections or Slack history for context on an anomaly.
  • Render a chart or write a CSV when the answer is best shown visually.
The Activity Feed on the right shows each step as it happens. Every step is expandable so you can inspect the exact SQL, parameters, and raw output.

​
Artifacts

Anything non-conversational Datost produces — a Recharts chart, a large table, a CSV export, a generated notebook — becomes an artifact attached to the run.

Persistent

Artifacts live past the stream. Reopen a month-old Context and every chart still renders.

Downloadable

Click any artifact to download the underlying file — CSV, PNG, or notebook.

​
Sharing and persistence

Contexts are organization-scoped. Any teammate with access to the same data sources can open a Context URL (/context/{id}) and see the full history, artifacts, and reasoning. Rename, archive, or delete from the sidebar; archived Contexts collapse into a separate section but remain searchable.

​
Power-user tips

  • Enter sends; Shift+Enter inserts a newline.
  • Paste or drag images into the composer to give Datost a screenshot to reason about — useful for debugging dashboards.
  • Click any Activity step to expand its raw tool input and output.
  • Use New Chat to start a fresh Context when switching topics — shorter histories keep the agent focused and faster.
  • If a reply is going the wrong direction, hit Stop and rephrase — the partial run is preserved in history.