Your analyst has a 4-week backlog.
Ward answers in seconds.

Most ad-hoc questions take a week. Ward queries every data source at once and returns the root cause with charts the same day. Click any number to see the SQL.

Get a demo →

One question. Full investigation.

“Why did margin drop in the Southeast?” usually means three systems and a deck next week. Ward answers in seconds, across every connected source, with root causes, confidence intervals, and charts.

app.getward.ai

A question mid-answer in Ward chat, asking which stores missed their margin target last month and what is driving it at Corpus Christi, with the reasoning trace showing 7.6 seconds of thought, models listed in 8 milliseconds, and two models inspected in 16 milliseconds. — **Reasoning.** You watch it choose the models before it writes a word of the answer.

Ward’s settled answer: all six stores clustered between 31.2 and 31.7 per cent against a 32 per cent margin target, then the cause at Corpus Christi, where Electronics runs at 12 per cent margin with $389K of promotional lines sold below cost carrying $162K in discounts, followed by a suggested next question. — **The answer.** Six stores, one cause, and the follow-up worth asking. One question, seven round-trips, 87 seconds.

Ward’s tool inspector open on the query behind a figure, showing the two rows it returned and the JSON arguments that produced them: city equals Corpus Christi, department equals Electronics, business_date between 1 and 31 December 2025. — **Tool inspector.** Every number opens into the call that produced it: arguments, rows, raw result.

The per-message cost tooltip in Ward chat: 96,299 input tokens, 10,674 output, 106,973 total, roughly 123 tokens per second across 7 round-trips, estimated cost about $0.18. — **Cost.** Eighteen cents for the whole answer, priced per message, spent against your own provider keys.

Ward · one question about December margin 139s · one take

One question, seven round-trips, every query inspectable on the way past. The answer settles at 87 seconds. 139 seconds, unedited, one take, real time. Nothing is sped up.

Every answer ships with its evidence trail

Every finding shows its work: the forecast model, the MAPE, the confidence interval, the source tables. Pin findings, share them, or drill deeper. Any insight traces back to the row.

Audit the query. Audit the math.

See exactly what Ward ran: the raw SQL, the chart spec, the result set, the forecast model and its parameters. Export any query to your BI tool or hand it to engineering.

Charters on file for every agent

Each agent runs under a written charter: scope, sources, allowed actions, owner, version. Every edit is logged with name, time, and approver. Diff two versions in the console.

Powered by your choice of LLM

OpenAI, Anthropic, Gemini, or bring your own. Ward’s router picks the model per task and you can switch any time. Forecasting math runs on classical models, not the LLM.

Fixes powered by Chat