Plain English — with a cost gate and a sealed transcript.
One chat drives every product — datasets, synthesis, SCADA, ICS, agent runs. It plans first, asks for approval, streams the execution.
Nothing charges your account until you approve the plan, and the whole conversation seals into the same evidence chain as the rest of the stack.
What would you like to generate?
- 43
- tools
- 5
- families
- per-step
- cost gate
- sealed
- transcript
A driver, not a chatbot.
Six behaviours the assistant keeps on every run — so a non-engineer can drive the platform and a procurement lead can trust the spend.
Plans first, executes second
Every multi-step task starts with an explicit plan card. No blind tool calls — you approve before any credit is spent.
Cost-gated by default
Each step carries a credit estimate; the assistant pauses when a single step exceeds your soft cap.
Live execution stream
Tool runs stream in real time — start, progress, output preview and per-step duration.
Cancel and refine, mid-run
Stop a plan part-way, refine the prompt, replay from the last checkpoint. No work re-done unnecessarily.
Same evidence chain
Every artefact joins the same cryptographic hash-chained ledger as Fabricate, Synthesis, SCADA and ICS.
Replayable transcripts
Every conversation seals into a transcript artefact: prompt, plan, approvals, tool I/O, sealed bundle.
Review the plan below and approve to start execution.
It shows the plan and the cost — before it spends.
Ask in English. The assistant returns a plan card with the concrete steps, the engine it will use, and a credit estimate in plain numbers. Nothing charges your account until you tap approve — and every step, approval and tool call lands in the sealed transcript.
- Concrete steps + credit estimate up front
- Approve, refine or reject — you decide
- The whole exchange is sealed and replayable
The same tools, reachable in plain English.
The same tools the Agentic Data Scientist uses — reachable in plain English or directly from the SDK.
Data preparation
6 toolsMachine learning
7 toolsExplain & report
8 toolsSynthetic data
5 toolsIndustrial / ICS
6 toolsThe questions procurement asks.
Which model drives it?
Auto-routed: a small open model proposes the plan, a larger model only fires when reasoning depth justifies the cost. You never pick the model directly.
Where is my data sent?
The Assistant is powered by a third-party language model. Depending on how your workspace is configured, your request text, recent session history, dataset and column names, row counts, and — after a connector step — a small row preview are sent to the configured provider (OpenAI, Anthropic or Google Gemini). Your full datasets and generated artifacts are not. You can register your own provider key, and an operator can disable live model calls entirely; both paths fail closed. Deterministic generation, validation and evidence run inside RadMah's own infrastructure with no model provider involved.
Can I disable it per role?
Yes — RBAC scopes (read / generate / admin) gate which tool families a role can invoke. The audit log records every call.
What if a step fails?
The assistant retries deterministically inside its budget, then surfaces the exact failure with a recommended fix. No silent fallback.
Describe the job. Approve the plan. Keep the transcript.
In a working session we drive a real request end to end in plain English — you watch the plan, the cost gate and the live stream, and keep the sealed transcript.