Flagship · the agent runtime

48 wired modules. 43 typed tools. One sealed evidence chain.

The agent that runs the platform. It plans the work, asks before spending, executes typed Python tools, and self-heals when a quality gate trips.

Not a chatbot — a multi-step autonomous pipeline across every engine, with a human-approval gate on spend and the whole run sealed into one evidence bundle.

app.radmah.ai / agentlive
Agentic Data Scientist
Autonomous multi-step pipelines with human approval gates
+ New Project
Search by title or goal...
All51Active0Awaiting you1Complete21Failed29
blocked15 June 2026
A6w-ns
[Action panel] Direct execution of mock_data
Progress0/1 steps
Cancel Delete
Open
blocked15 June 2026
A6-ns
[Action panel] Direct execution of mock_data
Progress0/1 steps
Cancel Delete
Open
blocked15 June 2026
A5-ns
[Action panel] Direct execution of mock_data
Progress0/1 steps
Cancel Delete
Open
blocked14 June 2026
A4-noseal
[Action panel] Direct execution of mock_data
Progress0/1 steps
Cancel Delete
Open
complete14 June 2026
A2-seal
[Action panel] Direct execution of mock_data
Progress1/1 steps
Delete
Open
blocked14 June 2026
A2-noseal
[Action panel] Direct execution of mock_data
Progress0/1 steps
Cancel Delete
Open
48
modules
43
typed tools
6
engines
sealed
per run
◆ Module map

48 specialised modules. One coherent runtime.

Six families, each with one job. The 16 below are the core control surface; the rest are wired end-to-end in code.

Planning & approval

4 modules
AP

Agent Planner

Decomposes intent into a typed step graph; estimates cost per node before any tool fires.

PE

Plan Executor

Runs the graph with structured concurrency, retries and per-step checkpointing.

HAG

Human Approval Gate

Pauses on high-cost or high-risk steps; surfaces a typed approval card to the operator.

Sh

Shadow

Dry-runs proposed plans against last week's data to surface regressions before commit.

Cost & scheduling

2 modules
CM

Cost Manager

Hard caps per project, soft caps per turn; halts when burn rate exceeds policy.

ES

Engine-Level Scheduler

Routes work to the cheapest engine that satisfies the contract (Mock vs Synthesize vs SCADA).

Physics validation

1 module
PV

Physics Validator

Constrains generation to physically-realisable signal shapes with a physics-honest response model.

Self-heal & resilience

5 modules
SHX

Self-Healing Execution

Watches the evidence chain; if a quality gate fails, repairs the artefact and replays downstream.

RP

Replan

Mid-run plan revision when an unexpected error or insight makes the original plan stale.

REG

Risk Evidence Graph

Risk-stratified evidence graph; flags brittle or low-confidence sub-paths for review.

CCR

Cross-Correlation Rebuilder

Preserves PK/FK and joint distributions when artefacts are regenerated.

CAB

Consistency-Aware Builder

Re-derives downstream artefacts when upstream inputs change so the chain stays consistent.

Code generation & review

2 modules
CGC

Code-Gen Critic

Reviews tool-generated Python before execution; rejects unsafe or undefined behaviour.

CCM

Code Change Manager

Versions every tool-generated Python; lets the operator diff and roll back at any step.

Evidence & narrative

2 modules
PP

Proof Packets

Bundles per-step inputs, outputs, hashes and parameters into the signed evidence chain.

NG

Narrative Generator

Produces a plain-English run-log alongside the JSON artefacts — for the audit reviewer.

Also wired: approval_timeout · contract_context · dataset_resolver · evidence_parser · executor_eval · grounded_output · hypothesis_tree · memory · moderation · observer · patch_planner · policy_adapter · prompt_registry · replan_evaluator · replay_bundle · retry_escalate · rseg_calibration · scenario_selector · turn_context · validation_runner · budget · hagp_pause · surrogate · diff_builder · domain_prompt · … and more

◆ Tool palette

43 typed tools the agent can wield.

All five families run through both the chat interface and POST /v1/agent/tools/{name}/execute for direct invocation from the SDK or your own UI.

Data preparation

6 tools
data_clean
feature_engineer
transform_data
encode_categorical
scale_numeric
split_dataset

Machine learning

7 tools
train_predictive
tune_hyperparameters
cross_validate
evaluate_model
cluster
detect_anomalies
dimensionality_reduce

Explain & report

8 tools
shap_explain
feature_importance
profile_dataset
scatter_matrix
time_series_plot
decompose_timeseries
forecast
generate_report

Synthetic data

5 tools
mock_data
synthesize
fabricate_dataset
verify_synthesis
quality_report

Industrial / ICS

6 tools
scada_sim
ics_attack
inject_mock
analyze
verify
evidence_seal
◆ Self-healing execution

Quality gate fails? It heals itself.

When a divergence widens, a constraint slips or a hash mismatches, the self-healing layer isolates the failing sub-graph, repairs the artefact, and rebuilds the chain — without losing what already passed. The narrative file explains, in English, what happened and why the rebuild is sound.

01

Sense

Self-healing execution watches every quality gate, artefact hash and chain segment in real time.

02

Diagnose

The risk-evidence graph isolates the failing sub-graph; the rebuilder identifies which downstream artefacts are now stale.

03

Repair

Replan rewrites the affected sub-plan; the code-gen critic reviews any generated code before re-execution.

04

Re-seal

The consistency-aware builder rebuilds the chain; proof packets commit the new root; the narrative generator appends to the run-log.

app.radmah.ai / agentlive
← All projects
Autonomous Agent · Complete
1 credit est.● Complete

Water treatment pump station SCADA, 60 seconds, Modbus + OPC-UA, 8 signals: discharge pressure, suction pressure, motor temperature, VFD output, flow rate, tank level, and an over-pressure alarm. [Protocols: modbus, opcua] [Duration: 1m] [Poll: 1 Hz] [Streaming: off]

small_wwtp
Progress2/2 steps
Execution steps
Step 1SCADA SimCompleted

Run a SCADA simulation for the water treatment pump station.

Sealedrealism: 75%KS: 0.83
Status: succeeded
Job: scada_sim #5f07d7
View evidence bundle
Started 9:07:58 AM · Completed 9:08:35 AM
Step 2VerifyCompleted

Verify the quality and provenance of the simulated data.

Sealedrealism: 80%
Status: succeeded
Job: verify #f4c7f5
View evidence bundle
Started 9:08:39 AM · Completed 9:08:47 AM
◆ One orchestrator

Every engine under one orchestrator.

The scheduler routes work to the cheapest engine that satisfies the contract; the physics validator keeps generation realisable; proof packets seal the result. The team that writes the contract sees the plan card, the cost gate, the quality verdict and the sealed bundle — nothing locked behind a platform admin.

Mock Data

Sub-minute deterministic synth from a prompt.

Synthesize

The trained synthesis engine on uploaded data.

Virtual SCADA

Physics-honest telemetry over real protocols.

ICS Security

Ground-truth ATT&CK ICS attack injection.

Tool executor

43 typed Python tools, in-process or queued.

Connectors

14 encrypted source adapters, browse + import.

◆ Posture

Built so an autonomous agent survives procurement.

Tenant-isolated

Every plan, prompt, dataset preview and tool I/O lives in your tenant prefix only.

Spend-bounded

Soft + hard cost caps; the agent stops, asks or downgrades the engine before exceeding policy.

Audit-ready

A sealed transcript per project — prompt, plan, approvals, tool I/O, evidence root.

Reproducible

Re-run the same contract + seed on a different cluster: the dataset hash matches byte-for-byte.

Bring a real prompt. Watch every module work.

A 90-minute working session: bring a representative dataset and one open question. We drive the agent end to end — you watch the plan card, the cost gate, the live stream and the self-heal trigger. You keep the sealed transcript bundle.