High-risk tools in Nitrostack

Accept one unexpired offer returned by request_compute. Only this explicit second step may materialize a live environment, and an offer can be accepted at most once.

FAST PATH: after record_consent, run the entire post-consent chain in ONE call — verify_kyc, screen_fraud,

Vision model checks uploaded photos against declared type. Flags contamination, rust, degradation. Outputs grade A/B/C with advisory flags. Runs autonomously on listing.created ...

Analyze uploaded video metadata, extract frames, run YOLOv8 object detection, and store detected events

Approve a queued high-risk action and execute it against live state. Use the action_id returned by the originating tool call.

Runs a quality audit across all pizza shops.

Route a tool call through Sentinel Gateway\

Chat with the Foundry Architect to get help with code.

Evaluate a request against the enterprise policy tree and make a definitive decision.

Pings a host (default 8.8.8.8) to check network latency and packet loss — a quick way to tell if lag is a network problem.. CRITICAL: Never output

Compute net income, proposed EMI, FOIR and DTI from the pulled bureau + bank data. Run after pull_bureau and fetch_bank_statements.

Mark a node unschedulable. SAFETY-ENFORCED: refuses if it would strand all workloads.

Evaluate evidence fusion and issue policy decision for transaction

Initiates a progressive canary deployment for a microservice using Argo Rollouts-style traffic splitting. HIGH RISK — queued for human approval. Automatically advances traffic f...

Locks the robot movement state immediately (E-Stop) or clears the lock to resume AUTO mode.

Long-running MCP task to orchestrate end-to-end clinical consultation evaluation.

Evaluates state event parameters for Decision Twin multi-agent workflow and returns agent blackboard state and proposal.

Alias for evaluateTransaction: Executes full multi-factor TBML evaluation pipeline.

Runs the full autonomous logic-evolution loop for an environmental shift as a long-running

Start the approved plan in safe simulation mode and monitor controlled execution.

Execute an argument vector inside an active environment owned by the authenticated agent. No shell is inserted; pass the executable as argv[0]. Output and runtime are bounded.

Plans a multi-waypoint path and executes a safe robot trajectory with CBF interception.

Plans a multi-waypoint path and executes a safe robot trajectory with step-by-step Control Barrier Function (CBF) interception.

Execute a safe Kubernetes rollback to the previous stable version.

Generate executable AWS CLI or Terraform commands to fix a specific security finding

Flushes the local DNS resolver cache — fixes sites that

Starts a blank project for collaborative workspace

Step 1: Dynamically geocode area/neighborhood names (e.g.

Perform GitHub operations

Send an email using the configured Gmail account.

Main AIDE router. Takes a natural language request and decides whether to schedule a meeting, assign a task, or run admin checks. Returns a structured RouterOutput.

Manually reports a new environmental shift for the swarm to react to (for demo/testing). Adds it to the active shift list so evolve_logic or run_mutation_cycle can then resolve it.

Full AI procurement pipeline. For each material: progressive radius search (10→25→50→75→100km), weighted supplier scoring (distance + price + grade + trust + quantity), industri...

Quarantines a network node suspected of compromise. HIGH RISK — queued for human approval. Confirm the node is not serving live traffic first.

Perform Kubernetes operations

Build an ordered plan for a life change: starting a first job, getting married,

Login with email and password

Main entry point for multi-agent clinical decision support graph. Parses transcript, triggers specialized agents, and compiles live clinical briefing.

Escalates a codebase branch for immediate compute priority and human review. HIGH RISK — queued for human approval.

Execute a raw SQL query against the Neon PostgreSQL database to retrieve sensor logs and historical data. Returns up to 100 rows.

Triggers a re-sync across all platform channels

Apply the automated fix, enforce S3 bucket encryption/access blocks, and resolve the GitHub P1 ticket automatically.

Restarts the Windows Explorer process (windows.exe / taskbar / file explorer). Fixes a frozen taskbar, unresponsive Start menu, or ghost icons without a full reboot — a classic ...

Retry a failed notification.

Roll a deployment back to its previous revision (equivalent to kubectl rollout undo).

Runs ALL 8 safety checks (drug interactions, allergy, disease, age, renal, pregnancy, duplicate therapy, and AYUSH herb-drug) in a single call. Returns pre-formatted check resul...

Run a simulated attack scenario to demonstrate Sentinel Gateway\

Execute the provenance-guarded red-team attack loop.\n\n

Launch a full autonomous red-team campaign. Evaluates mutated prompts in a continuous loop.

Runs a full Decision Twin negotiation cycle for a raw manufacturing event object. Unlike run_decision_twin_orchestrator (flat params), this accepts the event object directly and...

Triggers the multi-agent Decision Twin orchestrator for manufacturing & Industry 4.0 anomalies. Dynamically activates Planner, Plant Manager, Sensor, Maintenance, Memory, Produc...

Run the deterministic multi-agent fraud pipeline: ingest a ticket, generate Agent 1 triage, fan out to Agent 2 assignment and Agent 3 legal guidance, then return a dashboard-rea...

Run the complete demo sequence: setup servers → configure policies → normal traffic → tool poisoning attack → RBAC violation. Shows the full security pipeline in action.

Execute full audit: scan inventory, calculate waste, generate GitHub issue, and post Slack alert card in a single atomic step

Master Orchestration Tool: Executes the complete end-to-end SUMO traffic simulation pipeline in a single call (downloads network, generates routes, opens GUI, and analyzes resul...

Run the full triage pipeline on raw lab report text in one call: parse, classify, route to specialists, and produce a shareable summary.

Step 3b: Open the visual SUMO GUI app on your desktop to view vehicles driving on the map in real-time.

Step 3: Execute SUMO simulation headlessly using mymap.sumocfg and output stats.xml and tripinfo.xml.

Runs a Safe Test-Driven Mutation Cycle for a detected environmental shift: generates several

Triggers the full end-to-end multi-agent orchestration pipeline

Scale a deployment. SAFETY-ENFORCED: server re-reads the live PDB and policy floor and REJECTS

Runs all detection rules and returns ranked, evidence-backed authorization findings (BOLA, missing auth,

Runs pharmacogenomics + drug-interaction analysis through the full Secure Data Gateway pipeline (auth, RBAC, rate limiting, data-minimized AI Gateway routing, audit logging). Re...

Proactively heals a degraded fleet unit by rerouting its operational logic (e.g. falling back

Switches the active Windows power plan.

Changes the OS scheduling priority of a running process (e.g., set a game to High priority for smoother performance, or a background app to Low so it stops competing for CPU). A...

Run counterfactual simulation for candidate maintenance actions

Change live hospital conditions, then recalculate every dashboard and plan. No IDs required.

Run the full 4-step clinical workflow pipeline.

Run the complete persistent surge demo: change hospital conditions, compare plans, check policy, approve, execute and inspect the audit trail.

Test the target model v1 (baseline) with an adversarial prompt.

Test the target model v2 (patched) with an adversarial prompt.

Start tracking an approved and notified FactoryBrain plan.

Validates a manufacturing heuristic mathematically against sensor datasets. Exposes the rule and data to the Orchestrator LLM to compute statistical significance.

Verify project code