Now accepting design partners — 3 of 5 spots filled

No agent acts
without authority.

Edictum intercepts every AI agent action before execution — enforcing your policies, detecting prompt injections, and requiring human approval for high-risk operations.

SOC2 / ISO 27001 compatible On-prem or cloud deploy Works with Claude, GPT-4o, OSS 3-line SDK integration
edictum agent-proxy — live event stream ● LIVE
09:14:02.103 agent-prod-01 bash_exec ls -la /var/app/config/ ALLOW
09:14:03.847 agent-prod-01 file_read /var/app/config/database.yml ALLOW
09:14:05.221 agent-prod-02 bash_exec kubectl apply -f deployment.yaml PENDING ●
09:14:05.229 ⚠ GATE kubectl apply matches policy: require-human-approval-for-prod-deploy
09:14:09.003 agent-prod-03 file_write /etc/cron.d/backup [INJECTION DETECTED] DENY
09:14:09.011 ✕ BLOCK Prompt injection pattern matched. Action terminated. Session flagged.
09:14:12.778 human:j.marsh approved kubectl apply — agent-prod-02 deployment authorized ALLOW
09:14:13.001 audit session a7f2c1 — 4 allow, 1 pending→approved, 1 deny (injection). Immutable record stored.

The real incident that built Edictum: A Claude AI agent began deleting production files autonomously. The SOC team caught it — but only after the damage occurred. No intercept layer existed. No policy was enforced. No audit trail was left. That is the problem Edictum solves.

Four layers of agent control

Every AI agent action passes through Edictum's enforcement stack before it reaches your infrastructure.

01

Proxy Intercept

Transparent proxy sits between agent code and LLM API. Zero code changes required — point your SDK to Edictum.

02

Injection Scanner

Every message flowing into the agent is scanned for prompt injection patterns before the LLM ever sees it.

03

Policy Engine

YAML-defined rules evaluate every tool call in real time. Allow, deny, or escalate to human approval in <20ms.

04

Audit Trail

Append-only session log with full forensic replay. Tamper-evident evidence packages for SOC2 / ISO 27001.

Policy as code

Define agent governance in YAML. Policies are version-controlled, auditable, and enforced in real time — no code changes, no deployments.

  • Allow / deny by tool name, agent ID, resource path
  • Require human approval for irreversible actions
  • Rate limits, time-of-day restrictions
  • Alert on anomaly (baseline deviation)
policies/prod-agents.yaml ✓ active
# Edictum policy — production agents
agent: agent-prod-*

rules:
  - action: file_write
    path: /etc/**
    verdict: deny

  - action: bash_exec
    pattern: kubectl apply|helm upgrade
    verdict: require_approval
    approvers: [soc-team]

  - action: bash_exec
    pattern: rm -rf
    verdict: deny

  - action: *
    verdict: allow
    log: true

Fits into your existing stack

🤖 Anthropic
🔮 OpenAI
🐙 GitHub Actions
💬 Slack
🔔 PagerDuty
📊 SIEM / Splunk
☸️ Kubernetes
🪣 AWS S3
🔗 Webhook
📋 Jira
📂 LangChain
+ Any LLM

Integrate in 3 lines. No SDK changes required.

# Before
client = Anthropic()

# After — full governance enabled
from edictum import EdictumProxy
client = Anthropic(
  base_url=EdictumProxy(policy="prod-agents.yaml")
)

From people who needed this yesterday

★★★★★

"We had three Claude agents running in our CI/CD pipeline with zero visibility. Our CISO was furious when she found out. Edictum is exactly the control plane we needed."

JM

Jennifer Marsh

CISO, Series B SaaS

★★★★★

"The approval gate feature alone is worth it. My SOC team was getting paged at 2am because an agent tried to run kubectl commands in production. That's not acceptable."

KR

Kevin Reyes

SOC Manager, DevOps Platform Co.

★★★★★

"We're building AI agents for code review and deployment automation. Edictum gives us the audit trail our compliance team requires — and the block capability our AppSec team trusts."

AT

Aisha Thompson

AppSec Manager, Enterprise Software Co.

Simple, agent-based pricing

Design partners receive 40% off listed rate and direct input into the product roadmap.

Design partner

Team

$1–4K/mo

Up to 20 agents · unlimited events

  • Proxy intercept + policy engine
  • Human approval gates
  • Slack + PagerDuty alerts
  • 90-day audit log
  • SOC dashboard
Join as design partner

Includes 90-day pilot — no commitment required

Enterprise

$8–12K/mo

Unlimited agents · on-prem option

  • Everything in Team
  • SIEM / Splunk integration
  • SOC2 / ISO evidence packages
  • On-premise deployment
  • Dedicated security engineer
Contact us

Custom contract · SLA included

2 spots remaining

Design partners get 40% off, direct roadmap input, and a dedicated line to the founder. We only accept teams already running AI agents in production.

No spam. No sales calls. We'll review your use case and respond within 24 hours.