Skip to content
FDAI Forward Deployed Agents
Documentation

FORWARD DEPLOYED AGENTS

Cloud operations,
from evidence to action.

FDAI brings evidence, policy, and accountable agents together to manage cloud change, reliability, and cost. Begin by observing; allow changes only within authorized boundaries.
Get started How it works

Azure-native. Evidence-led. Human-authorized.

Open Neural View

OPERATIONAL FOCUS

Three priorities. One operating model.

FDAI connects the evidence behind a change, a reliability concern, or a cost signal with a governed next step.

Change Safety

Understand a change before it runs.

Review proposed changes against policy, impact scope, and rollback readiness. Keep the decision and its evidence together.

Explore change safety Resilience

Make recovery a practiced capability.

Connect incidents with their dependencies and evidence. Use bounded recovery actions and drills to assess readiness.

Explore resilience Cost Governance

Give every cost decision context.

Investigate cost anomalies and right-sizing opportunities. Review the evidence before an eligible change is allowed to run.

Explore cost governance

Available actions depend on connected evidence, configured policies, and each action's promotion state.

HOW IT WORKS

Rules before reasoning. Evidence before action.

One operating loop keeps context, decision, and outcome connected. A language model is a tool for ambiguity, not a source of permission.

A clear path from signal to outcome.

  1. Evidence

    Signals grounded in scope, dependencies, time, and provenance.

  2. Decision

    Rules and patterns first. Residual reasoning needs cross-checks and a verifier.

  3. Authorized action

    Policy, safeguards, and applicable approval.

  4. Verified outcome

    An independent observation, recorded in the audit trail.

Conceptual flow, not live activity.

Reasoning is not permission. An accepted request is not a verified outcome.

Explore the architecture

AUTONOMY WITH ACCOUNTABILITY

Start by observing. Expand with evidence.

New actions begin in observation mode. They record proposed decisions without changing resources; promotion is a separate, evidence-based decision.

Observation and promotionMeet the 15 agentsUnderstand typed actions

  • Human authority stays explicit

    Changes require current approval or valid standing human authorization where applicable. Silence is never approval, and approval and executor identities stay separate.

  • Responsibilities stay distinct

    Fifteen specialized agents own observation, judgment, approval delivery, execution, and audit. No agent can approve its own work.

  • Uncertainty stays visible

    Insufficient evidence leads to a hold, denial, recovery, or human review with an audit record. Deployment alone never grants permission to act.

Seven safeguards before an autonomous change

Missing any one blocks execution. Approval and independent effect verification remain separate requirements.

  1. A clear stop conditionKnow when to halt instead of continuing.
  2. A tested rollbackKeep a rehearsed recovery path.
  3. An impact-scope limitBound the resources a change can affect.
  4. A successful dry-runCheck the proposed effect before applying it.
  5. A logical-target lockPrevent conflicting changes to the same target.
  6. A stable idempotency keyMake retries safe without duplicate effects.
  7. A two-phase auditRecord intent before action and its terminal result.

YOUR NEXT STEP

Begin with one well-defined scope.

An accountable owner, reliable evidence, and a baseline come before broader automation. Choose the guide that fits where you are.

Understand FDAI

A short orientation to the product, its fit, and its limits.

Get started

Plan a pilot

Review deployment prerequisites and prepare a bounded Azure scope.

Read the deployment guide

Explore the design

See the agents, evidence boundaries, and operating architecture.

Read the architecture

Azure is the implemented target. Results depend on your scope and measured evidence; no performance improvement is assumed.