Automedon v1 · Platform record

This page documents the first Automedon platform, preserved as built. For the company and current work, see Automedon Technologies.

Active scenario

Human-GuidedAutonomy

Operators maintain control while autonomous agents execute at scale.

Explore the platform Request access
Why mission essential

Speed, control and accountability. Not two of three.

The gap between traditional automation and modern operational requirements demands a new approach.

Traditional

Scripted Automation

  • Rigid, pre-defined workflows
  • Breaks on unexpected conditions
  • No adaptive decision-making
  • Single point of failure
  • Manual intervention required
Alternative

Autonomous AI

  • Black-box decision making
  • Unpredictable behavior
  • No accountability chain
  • Difficult to audit
  • Trust without verification
Automedon

Human-Guided Autonomy

  • Adaptive mission planning
  • Real-time human oversight
  • Full decision provenance
  • Multi-agent coordination
  • Speed + Control + Accountability
10× Faster than manual operations
100% Human oversight maintained
0 Black-box decisions
Technology stack

Seven integrated layers

From operator intent to tool execution, every layer is built for mission-critical autonomous operations.

Mission Planner

What it does

Translates high-level objectives into executable multi-step plans. Decomposes complex missions into coordinated agent tasks with dependency management.

Why it matters

Operators define intent, not implementation. The planner handles tactical decomposition while maintaining strategic alignment.

What others don't have

Dynamic replanning on failure. If an agent hits an obstacle, the planner adapts without operator intervention.

Orchestrator

What it does

Real-time coordination of multiple AI agents. Manages task assignment, resource allocation, inter-agent communication, and execution timing.

Why it matters

Single agents are limited. Coordinated swarms multiply capability. The orchestrator is the command layer that makes multi-agent operations coherent.

What others don't have

1 Hz heartbeat monitoring with automatic failover. Agents that go dark are detected and replaced within seconds.

Agent Pool

What it does

Specialized AI agents for different mission profiles: offensive operations, defensive monitoring, research analysis, forensics, and incident response.

Why it matters

Different tasks require different capabilities. Purpose-built agents outperform generalists on specialized missions.

What others don't have

LLM-agnostic architecture. Deploy Claude, GPT, or local models based on classification requirements. No vendor lock-in.

Tool Adapters

What it does

Standardized interfaces to operational tools: osquery, YARA, Sigma, Volatility, tshark, Sysmon, and custom integrations.

Why it matters

Agents don't just think, they act. Tool adapters bridge AI reasoning to real-world execution with validated outputs.

What others don't have

Structured output validation. Tool results are parsed, verified, and formatted before agent consumption. No hallucinated tool outputs.

Explainability Engine

What it does

Real-time reasoning transparency. Operators can query any agent mid-execution: "Why did you choose this approach?" and get immediate explanations.

Why it matters

Trust requires understanding. In defense contexts, operators must know why an agent took an action, not just what it did.

What others don't have

Interactive steering. Don't just observe, intervene. Adjust agent behavior mid-mission without stopping execution.

Safety Layer

What it does

Multi-level approval system: Allow once, Allow for session, Allow always, Deny. Silent-command watchdog detects unauthorized execution attempts.

Why it matters

Autonomous systems must have hard limits. The safety layer ensures agents operate within defined boundaries with human approval gates.

What others don't have

Policy enforcement with cryptographic verification. Approval decisions are logged, signed, and auditable.

Decision Provenance

What it does

Complete audit trail of every decision, action, and outcome. Traceable chain from operator intent through agent execution to mission result.

Why it matters

Accountability requires records. Post-mission analysis, compliance audits, and continuous improvement all depend on decision provenance.

What others don't have

Mission replay capability. Review any operation step-by-step with full context reconstruction.

Live explainability

Operators don't just watch. They query and steer.

Every agent can be interrogated mid-mission and redirected without stopping execution.

Operator console Session 2F41 Live
operator@automedon:~$ explain AGENT-02 last_decision
AGENT-02 Decision Trace Selected lateral movement detection playbook based on:
  • Observed SMB traffic spike (147% above baseline)
  • Multiple failed auth attempts from single source
  • Pattern matches Sigma rule "win_susp_psexec"
Confidence: 94% | Alternative considered: Brute force (rejected, no password spray pattern)
operator@automedon:~$ steer AGENT-02 --add-focus "credential_dumping"
✓ AGENT-02 focus updated. Now prioritizing credential dumping indicators.
operator@automedon:~$

Query any decision

Ask agents why they chose a specific approach. Get structured reasoning traces, not vague summaries.

Mid-mission steering

Adjust agent focus without stopping execution. Add constraints, shift priorities, or redirect attention.

Confidence visibility

See uncertainty levels for every decision. Agents that aren't sure say so, and explain why.

Safety and control

Every action is gated, logged and reversible

Autonomy without accountability is liability.

Approval gates

Four-tier permission system

Sensitive operations wait for a human decision. Each decision is scoped, signed and logged.

OnceSingle execution approval
SessionValid for current mission
AlwaysPermanent allowlist
DenyBlock and log attempt
Watchdog

Silent command watchdog

Monitors for unauthorized execution attempts. If an agent tries to bypass approval gates or execute unregistered commands, the watchdog intervenes immediately.

Active 0 violations detected
Policy enforcement

Boundaries set before the mission

Policies specify allowed tools, forbidden actions, escalation thresholds, and automatic halt conditions.

policy: no_destructive_ops escalate_on: privilege_change halt_if: network_exfil > 10MB
Rollback

Instant rollback

Every agent action is reversible. If a mission goes sideways, operators can roll back to any checkpoint with full state restoration.

System architecture

Operator intent to mission result

End-to-end visibility across every layer. Explainability and safety wrap the whole chain.

Explainability
Human layer
Operator Mission objectives and oversight
Intent and steering
Planning layer
Mission Planner Objective decomposition
Task plans
Execution layer
Orchestrator Real-time coordination
Task assignment
Agent layer
Agent-01
Agent-02
Agent-03
Agent-N
Tool execution
Tool layer
osquery YARA Sigma tshark Volatility Sysmon Custom
Safety and provenance
Our mission

Ensure the United States leads the cyber domain in the age of AI.

Automedon Technologies builds human-guided autonomous systems for defense and critical infrastructure. We believe the future of cyber operations requires AI that can act, not just advise, with humans firmly in control.

LLM Agnostic
24/7 Operations
100% Human oversight

USA

Access

Preserved, not sold

Automedon v1 is kept as a record of what we built and how it works. Access requests for current programs go through the company page and are reviewed by a person.

Request access Automedon Technologies