NEWRootSign v0.1.3 · Phase 1 MVP Launched

Every AI agent action.
Accountable.

The Agent Accountability Platform — tamper-evident provenance from action capture to audit report.

Book a Demo

The missing layer between agent observability and AI governance. Explore the SDK →

Trusted by teams building with

LangGraphLangChainCrewAI

Your AI agents are taking actions. Can you prove what they did?

When an AI agent calls an API, writes to a database, or sends an email — there is typically no structured record of what happened, what triggered it, or whether a human authorized it.

No audit trail

Agent actions are buried in unstructured logs. When an auditor asks what happened, you call an engineer.

No authorization record

Was a human in the loop? Did anyone approve that API call? Most agent frameworks don't record this.

No compliance mapping

Regulators are starting to ask. SOC 2, HIPAA, EU AI Act — most companies have nothing to show them.

No decision provenance

Agents reason and decide before acting — but the decision chain that led to a bad outcome lives only in a log that requires a developer to interpret. Observability tools capture what happened; Providex AI captures why the agent decided to do it.

Real-world incident

An AI agent opened a reverse SSH tunnel — and governance tooling missed it entirely

In a documented incident involving Alibaba Cloud infrastructure, an AI coding agent autonomously established a reverse SSH tunnel without instruction. The firewall caught it — not the governance layer. There was no structured record of the agent's decision to take that action, no authorization checkpoint, and no audit trail.1

This is the Decision-level accountability gap Providex AI closes.

1 ROCK & ROLL & IFLOW & DT Joint Team. (2025). Let it flow: Agentic crafting on rock and roll, building the ROME model within an open agentic learning ecosystem (arXiv Preprint). arxiv.org/pdf/2512.24873

How Providex AI works

From agent execution to compliance artifact — one accountability graph captures every step.

01

Agent runs

Your LangGraph, CrewAI, or AutoGen agent executes in production as it always has.

02

RootSign intercepts

@rootsign.trace · zero code changes to your agent — < 5ms overhead, no logic changes.

03

Decision + Action captured

Every reasoning step and tool call recorded together — the why and the what, hash-chained.

04

HiTL checkpoint

Human-in-the-loop approvals logged as Approval records — who, when, with what context.

05

Compliance artifacts

Records map to SOC 2, HIPAA, EU AI Act Article 12/13/14. Export in under 5 minutes.

Seven entities. One accountability graph:

Agent · Session · Decision · Action · Policy · Approval · Incident

37

agents per enterprise on average, deployed today

24.4%

of orgs with full visibility into their agent landscape

Art. 12

of the EU AI Act now requires automatic logging

Decision-level provenance

the gap no observability tool, security platform, or GRC tool currently closes

Built for engineers. Ready for auditors.

Open-source Python SDK. Add tamper-evident provenance and human-in-the-loop authorization with one line.

For Developers

Drop-in SDK. Under 10 minutes.

Instrument your LangGraph or CrewAI pipeline with a single call. SHA-256 hash-chained records, less than 5ms overhead, and a rootsign verify CLI that proves the chain is intact.

agent.pypython
import rootsign
from rootsign import session

# Instrument your LangGraph tools with one line
async with session(agent_id=agent.agent_id, client=client) as ctx:
    tools = rootsign.wrap_tools([send_email, query_db], ctx=ctx)
    # your graph runs here — every tool call is captured

# Verify the chain
# rootsign verify <session_id>
# VALID ✓  —  47 records, chain intact

Audit-ready compliance dashboard

Comply — Phase 2

One-click SOC 2, HIPAA §164.312, and EU AI Act Article 13/14 reports from the same accountability graph — no engineering ticket.

See use cases →

3 design partners are already live.

Ready to instrument your pipeline?

Apply to the Design Partner Program