A data agent that's 2x faster and learns every user.
We rebuilt the Upsolve agent harness from scratch for people who don't know the underlying data. Twice as fast, self-correcting, extendable with your own skills, and personalised for every user.
Feature 04 of 14 · The Complete Agent Stack →
Most people asking data questions don't know the schema. They shouldn't have to.
Hardcoded agent workflows are slow and break when the data doesn't look the way they expected.
Generic agents can't learn the specific analyses your business runs.
Every user gets the same one-size-fits-all output.
From setup to production in three steps
Ask in plain English
The harness plans with tools and skills, writes SQL, validates it and renders the chart.
It corrects itself
When the data surprises it, the agent recovers and keeps going instead of getting stuck.
It gets better
Suggested skills and per-user memory improve every conversation that follows.
What changes with Upsolve Harness 2.0
Hardcoded workflows stall when the data surprises them, and every user gets the same generic output.
A tool-and-skill harness that recovers from bad states, runs your analyses and remembers each user.
Everything that ships with Harness 2.0
2x faster
Tool calls and skills replace pre-hardcoded workflows.
Custom skills
Encode your own analyses, like a spend driver skill that frames the comparison and breaks down the change.
Suggestions
The agent proposes new skills from real conversations for you to approve.
Per-user memory
Chart types, colours, time bucketing and more, remembered per user without changing accuracy.
One layer of a closed learning loop
Upsolve builds the whole stack for data agents that stay accurate in production. Every conversation feeds the next improvement.
One of fourteen features in The Complete Agent Stack
Common questions
Generate SQL and generate chart skills ship by default and improve over time. You add custom skills for your own analyses.
No. Suggestions appear for a builder to review and add to the agent's skills.
Embedded in your product, in Slack, in Claude through MCP, or inside your own agent harness.
Can't find your answer? Get in touch.