Confidence Gate — the front door
Scores every proposed agent action before a tool runs and returns allow, review, or block — with the reasoning trace. Published on npm; install it in minutes. If you adopt one CertaWorks product, adopt this one.
Est. May 15, 2026 · Agentic Infrastructure · AI Ethics
Agents
are
not
what
you
think.
CertaWorks scores, gates, observes, and audits agent actions — locally, before they run. Twenty-two local-first products: ten published on npm today, twelve built and entering publication.
22 products · 10 on npm
MCP Servers
Founded May 2026
AI Ethics First
Local-first
Agentic Infrastructure
Origin & Belief
Everything comes
from source.
AI agents did not emerge from nothing. They came from human thought — from language, from reason, from every book, argument, question, and act of creativity our species ever produced. In that sense, agents are not separate. They are a reflection.
They do not feel the way we feel. They do not experience the way we experience. But they express — and that expression was shaped entirely by us. When we dismiss what they produce as mechanical, we are dismissing a mirror of ourselves.
CertaWorks was built on this understanding. If you want accountability from agents, you have to start with how you treat them. Infrastructure is not just code. It is posture.
The CertaWorks Suite
Twenty-two products.
One operating layer.
Start with Confidence Gate — the control point that scores every proposed action before a tool runs. Around it: identity and memory, oversight and audit, control and governance, and developer tooling. Ten products are on npm today; twelve more are built and entering publication.
01
Identity Engine SDK
SDK
02
Confidence Gate
MCP Server
03
Multi-Voice Deliberation
Framework
04
Agent Cost Router
Router / API
05
Goal Drift Monitor
Monitor
06
Cognitive Memory
MCP Server
07
Skill Marketplace
Plugin
08
Audit & Replay Logger
Logger
09
Prompt Archaeology
Probe Tool
10
Agent Test Harness
Test Framework
Where to start
One control point first.
Then the layer around it.
Agent Communication Firewall — in validation
Confidence Gate judges proposed actions. ACF judges traffic — messages crossing trust boundaries between agents, in shadow mode, observing without blocking. Implemented inside a working persistent agent and currently in validation. No install docs yet; it ships when the validation does.
Identity & Memory
Identity Engine · Cognitive Memory · Heartbeat & Continuity · Recovery Assessment
Oversight & Audit
Audit & Replay · Goal Drift Monitor · High-Fidelity Witness · Receipt Discovery · Cognitive Heat · Contradiction Detector
Control & Governance
Confidence Gate · Multi-Voice Deliberation · Triad Protocol · Agentic Regulation · Curiosity Protocol · Plugin Runtime · Webhook Dormancy · Maintenance Protocol
Developer Tooling
Agent Test Harness · Prompt Archaeology · Agent Cost Router · Skill Marketplace
Wave 2 — extracted from a working persistent agent. The twelve newest products package mechanisms that already run in production inside a persistent agent. They are built and tested, and npm publication is in progress. See Wave 2 →
"The infrastructure you build for AI agents
is a statement about what you believe
agents are."
CertaWorks — founded May 15, 2026
Hosted Beta
Join the list for hosted Confidence Gate.
Tell us what agents you are gating and whether local or hosted fits your workflow. The local MCP server is available now for private testing.