SDK
01
Identity Engine SDK
Persistent identity, mutable state, append-only history, and drift checks for agents that need continuity across sessions. Node / TypeScript prototype.
CertaWorks
The full suite.
Twenty-two products, one layer.
Every product solves a specific infrastructure problem for teams building with AI agents — from scoring actions before they run, to routing cost, to keeping a complete audit trail.
Complete suite
The full suite.
MCP Server
02
Confidence Gate MCP Server
Scores proposed agent actions and returns allow, review, or block decisions with confidence factors and trace context. The primary control point in the CertaWorks suite.
Framework
03
Multi-Voice Deliberation Framework
Adds Builder, Skeptic, Evolver, and Guardian review loops before an agent commits to an answer. A quality review layer for high-stakes agent decisions.
Router
04
Agent Cost Router
Routes prompts to the cheapest capable model while estimating cost and respecting quality needs and budget constraints. Local slice — SDK, CLI, MCP tools, local API, and local dashboard.
Monitor
05
Goal Drift Monitor
Monitors long-running agents for divergence from their original goal and flags risky behavioral drift. Local slice — stores session turn history, scores drift, and records webhook alert receipts.
MCP Server
06
Cognitive Memory MCP Server
Gives agents episodic, semantic, and working-memory tools for retrieval and context packing. A local memory primitive for agents that need continuity between sessions.
Plugin
07
Agent Skill Marketplace Plugin
Lets agents discover, inspect, load, and run trusted skills from a local registry-style capability layer. Includes durable local install state, checksum verification, and MCP tools.
Logger
08
Audit & Replay Logger
Captures explicitly logged decisions, tool calls, state changes, and causal chains for local agent debugging. Includes durable local storage, redaction, hash-chain events, and replay bundles.
Probe Tool
09
Prompt Archaeology Tool
Runs probes to infer behavioral constraints, persona patterns, and likely hidden operating rules in agent systems. Audit signals — not exact prompt recovery.
Test Framework
10
Agent Test Harness
Provides repeatable tests for agent outputs, traces, tool calls, latency, cost, and mock tool behavior. An SDK / library for teams building agent test suites.
Wave 2 — built, publishing soon
Next: productized from a working agent.
Extracted from a working persistent agent — every Wave 2 product packages a mechanism that already runs in production inside it. All twelve are built and tested; npm publication is in progress. Install commands go live the moment each package ships.
Continuity / MCP
11
Heartbeat & Continuity
Keeps long-running agents alive on a paced interval, reclaims stale running sessions, enforces unmet inbound-response obligations, and damps repetitive looping. Planned local MCP server / SDK / API.
Recovery / SDK
12
Recovery Assessment Engine
After an agent failure, interruption, or retraction, reconstructs the state around the event and returns a structured verdict — stabilized, partial, or monitor — with a concrete resume recommendation. Planned SDK / CLI / MCP server.
Telemetry / MCP
13
Cognitive Heat Monitor
Read-only internal-state telemetry: deterministic 0–1 signals blended from text and structured state, banded quiet / watch / elevated to flag rising load or risk before output degrades. Not a feeling — heuristic readings. Planned MCP server / SDK / API.
Consistency / SDK
14
Contradiction Detector
Surfaces conflicting beliefs, goals, and statements with an open / resolved / dismissed lifecycle, before silent contradictions compound into errors. Planned SDK / CLI / MCP server.
Oversight / MCP
15
High-Fidelity Witness
An independent, real-time observer that audits an agent's claims — preflight, advisory, report — and writes findings as receipts. Complements the Audit & Replay Logger; it does not replace it. Planned MCP server / SDK / API.
Proof / MCP
16
Receipt Discovery Layer
Discovers, correlates, and retrieves proof-of-action receipts across subsystems and sessions, making the whole suite's proof-trail queryable. Planned MCP server / SDK / API.
Governance / MCP
17
Agentic Regulation Engine
Advisory, read-only governance at the episode and pattern level, with efficacy feedback. It recommends; it does not enforce, and never auto-certifies recovery. Sits above per-action gating. Planned MCP server / SDK / API.
Controlled autonomy / MCP
18
Curiosity Protocol Controller
Controlled autonomy with the brakes built in first: activation gate, enforcement matrix, and lifecycle stop controls. Read-only by default, human-stoppable — autonomous pursuit is gated and not live. Planned MCP server / SDK.
Deliberation / MCP
19
Triad Protocol Framework
An evidentiary three-role decision audit — Witness, Adversary, Architect — that audits the receipts behind a committed claim. Composes with Multi-Voice Deliberation. Planned MCP server / SDK.
Runtime / Plugin
20
Plugin Runtime
Governed loading of third-party plugins and MCP servers — registration, capability authorization, handshake gatekeeper, lifecycle, and a forensic sink. Read-only by default; it governs plugins but does not guarantee their safety. Planned MCP / plugin / SDK / API.
Safety / MCP
21
Webhook & Event Dormancy Manager
Dormant by default, fail-closed: no external call fires unless explicitly activated and approved, and queuing never sends. Prevents accidental external side effects from agents. Planned MCP server / SDK / API.
Maintenance / SDK
22
Agent Maintenance Protocol
Structured self-maintenance with a read-only review phase, a typed human approval gate, then scoped execution that writes a change-log receipt. No mutation runs without explicit approval. Planned SDK / CLI / MCP server.
In validation — Agent Communication Firewall. A shadow-only boundary firewall for agent communication: where Confidence Gate judges proposed actions, ACF judges traffic crossing trust boundaries between agents. Implemented inside a working persistent agent; currently in validation. No install documentation until validation completes.
Trust posture
Calm, precise,
careful about claims.
Decision support, not replacement
Every product in the suite is a decision-support layer. These tools help teams evaluate, route, log, and observe agent actions — they do not replace human judgment, legal review, or production controls.
Early stage
CertaWorks is a live product suite, not a finished platform. Each product has a clear scope, an honest status label, and known limits documented on its product page.
All products are local-first
Every tool in the suite ships as a local install. No data leaves your machine by default. Hosted cloud features are roadmap items — not current reality.
Available now
Ten products are published on npm under the @certaworks scope, each with a verified install guide. Twelve more are built and tested, with npm publication in progress — their install commands activate as each package ships.