Natural data requests and follow-ups

analyze these numbers: 2, 4, 6
analyze this CSV:
name,value
A,10
B,20
show inputs
show steps
show raw results

When multiple CSV columns are numeric, HuLaHo asks which one to use. After a workflow, change an existing numeric input with change rate to 20%, or supply same with: followed by a JSON object. Compare this with the previous run reports scalar differences. These continuations execute real tools in the current session.

Start with a recipe

Sixteen ready-made workflows combine parsing, statistics, time series, classification, text processing and other calculations. Inputs shown by recipe help are demonstration values; replace them with your own values.

recipes
recipe help csv_statistics
recipe: csv_statistics {"text":"name,value\nA,10\nB,20","column":"value"}
save this workflow as csv_report
learned workflows

Plan and execute

Orchestrate accepts a JSON dependency graph. Whole-value references pass computed results into later tools. The executor orders dependencies, rejects cycles and unsupported tools, records failures and skips affected dependents. Only bounded pure computations are available. Natural-language planning requires an enabled local model; the generated plan is validated and gets at most one repair attempt.

orchestrate: parse this CSV and calculate the mean of value: name,value; A,10; B,20

Learn by successful execution

Successful parameterized plans can be retained in your private workflow collection while private learning is enabled. Reuse executes the same validated steps with supplied inputs; saving a procedure does not retrain neural weights. Use delete workflow NAME to remove it.

Inspect worked examples

worked example: business_case
hybrid next: the river
predict intent: what can you help me with

Worked examples rerun bundled synthetic inputs. Prediction scores estimate likely wording or an interaction category, not factual certainty.