Skip to main content
When you ask Tell a question with tell ask or through MCP, the response comes back as blocks — a streaming JSONL format that renders as rich UI on any platform. The same JSON produces charts in your terminal, on the web dashboard, and in native apps.

​
How it works

An LLM generates one JSON object per line. Each object has a t field that identifies its type. Your client reads lines as they arrive and renders each block immediately — no waiting for the full response. 
{"t":"text","c":"Here are your metrics:"}
{"t":"metric","title":"DAU","value":12847,"unit":"users","change":15.2}
{"t":"chart","v":"sparkline","data":[[0,10],[1,15],[2,12]]}
This renders as a text paragraph, a large number with a trend arrow, and a sparkline chart — all streaming in real time.

​
Block types

Tell supports 8 block types:
TypeWhat it shows
textPlain text paragraph
metricSingle KPI — value, unit, trend, color
metrics2-4 KPIs in a row
chartLine, bar, area, pie, or sparkline
tableRows and columns with headers
statusHealth indicator (healthy, degraded, down, unknown)
calloutAlert box (info, warning, error, success)
codeSource code with syntax highlighting

​
metric

A single number with optional context: 
{"t":"metric","title":"Error Rate","value":0.12,"unit":"%","change":-5.2,"color":"success"}
  • title — what the number measures
  • value — the number itself
  • unit — display unit (users, %, ms, etc.)
  • change — percentage trend (positive = up, negative = down)
  • color — hint: success, error, warning, info, or default

​
metrics

Multiple KPIs side by side: 
{"t":"metrics","items":[
  {"title":"DAU","value":12847,"unit":"users"},
  {"title":"WAU","value":45230,"unit":"users"},
  {"title":"MAU","value":128450,"unit":"users"}
]}

​
chart

Data visualization with five chart variants: 
{"t":"chart","v":"line","title":"User Growth","data":[[1704067200000,10000],[1704153600000,10500],[1704240000000,11200]]}
  • v — chart variant: line, bar, area, pie, or sparkline
  • data — array of [x, y] pairs (x is typically a timestamp in milliseconds)
In the TUI, sparklines render as compact Unicode bars (▂▃▅▆▇).

​
table

Structured data with headers: 
{"t":"table","cols":["Service","Requests","P99"],"rows":[["api",52000,"45ms"],["auth",31000,"23ms"]]}
Cell values can be strings or numbers.

​
status

Health indicators for systems or components: 
{"t":"status","label":"API Gateway","state":"healthy","message":"All operational"}
States: healthy (green), degraded (yellow), down (red), unknown (gray).

​
callout

Highlighted notices: 
{"t":"callout","v":"warning","title":"Rate Limit","c":"You're at 80% quota"}
Variants: info, warning, error, success.

​
code

Source code with optional syntax highlighting: 
{"t":"code","lang":"sql","c":"SELECT event_name, COUNT(*) FROM events_v1 GROUP BY event_name"}

​
Streaming

Blocks stream as JSONL — one complete JSON object per line. The parser buffers partial data and emits blocks as each line completes. If a line can’t be parsed, it’s reported as an error without crashing the stream. This means LLM output renders progressively as it’s generated.

​
Where blocks appear

  • tell ask — CLI responses rendered in the terminal
  • MCP tools — AI assistant responses in editors and chat interfaces
  • TUI — interactive dashboard views
  • Web dashboard — rendered as React components