CARC: Consequential Agent Runtime Control
Runtime control
for high-stakes AI decisions.
CARC gives teams deploying AI in high-stakes decisions an independent runtime control layer and audit trail for governing, reviewing, and reconstructing agent behaviour before outputs enter real-world workflows.
Keep your AI. Add runtime control. Send a question first.
01. Runtime Architecture
A control layer between your agent and the workflow it acts in.
CARC does not replace your AI models or agents. It runs alongside them, in the path between what your agent decides and what happens next in the workflow.
Every recommendation, escalation, deferral, and override passes through CARC's runtime engines, where it becomes observable, governable, and reconstructable.
Adapter Registry
Connects CARC to your agent's existing interfaces without requiring a rebuild.
Runtime Engine
Observes agent behaviour as it happens and routes it through policy and risk evaluation.
Evidence Engine
Captures decision context, human reviewer response, and outcome into a structured, reconstructable record.
Control Console
Where runtime sessions, agent runs, and human review come together for your team.
Agent
Registry
Engine
Engine
Console
02. Core Capabilities
Runtime control is
the missing layer.
Teams deploying AI in high-stakes decisions need to govern behaviour, not just documentation. CARC makes agent behaviour visible, owned, interruptible, and reconstructable while your product is operating in real workflows.
CARC's control layer sits between model approval and operational evidence. It shows when the agent acted, why it acted, what context it used, and who intervened.
Agent Behaviour
Know what the agent recommended, when confidence changed, when it escalated or deferred, and whether similar cases produce consistent behaviour.
Decision Context
Capture the decision context, workflow state, model version, thresholds, alerts, suppression rules, and intervention options that shaped the agent's action.
Human Intervention
Track who accepted, modified, overrode, escalated, or rejected the agent's recommendation, with evidence that can survive audit and incident review.
Can you reconstruct what the agent did, why it acted, what context it used, how the human reviewer responded, and whether the same case would behave the same way again?
03. Integration Models
Keep your AI. Add runtime control.
CARC integrates alongside the AI you already have. Choose the integration pattern that fits your architecture. No model rebuild required.
SDK-Embedded
Instrument your agent directly with the CARC Runtime SDK, wrapping recommendations, escalations, and overrides as they happen.
Adapter / Gateway
Route agent traffic through a CARC adapter sitting in front of your existing agent, with no changes to the agent's own code path.
Event Ingestion
Stream existing decision and intervention events into CARC's Runtime Engine for evidence capture and reconstruction without live instrumentation.
04. Developer Experience
Built for engineering teams, not just compliance teams.
Design Partner Preview
The CARC Developer Platform, Runtime API, and Control Console are implemented today and available to design partners while hosted public access rolls out.
Illustrative example
05. Runtime Metrics
Operational evidence for every high-stakes agent decision.
Runtime Metrics turns raw agent activity into evidence: what happened, who was involved, and whether it can be reconstructed later. Interface shown below is illustrative, not live customer data.
Sample interface data for illustration only
06. Use Cases
Built for teams deploying AI agents in high-stakes decisions.
For teams where AI agents are entering deployment, audit, post-market monitoring, or regulated use, and where behaviour after deployment must be controlled, not inferred.
Healthcare
Triage, diagnostic support, and care workflow agents preparing for deployment, audit, and post-market monitoring under MDR and the EU AI Act.
Hiring
Candidate screening and ranking agents where every shortlist, rejection, and recruiter override needs to be explainable and reviewable.
Lending
Credit decision and underwriting agents where approvals, declines, and overrides need a reconstructable record.
Insurance
Claims assessment, pricing, and risk review agents where escalations and human overrides must hold up in audit and disputes.
Public sector
Benefits, eligibility, and case-prioritisation agents where decisions about citizens need accountable human oversight and a reconstructable record.
07. Design Partner Program
Help shape runtime control for high-stakes AI.
We're working with a small group of teams deploying AI agents in consequential decisions. You bring one real agent workflow. We bring CARC and work with you directly.
Who it's for
Teams in healthcare, HR and hiring, lending, insurance and the public sector running (or about to run) an AI agent that influences decisions about people, especially high-risk systems under the EU AI Act.
What you get
- Free access to CARC during the pilot
- Founding-partner pricing after launch
- Direct input on the roadmap and policies CARC supports
- Evidence for human oversight under EU AI Act Article 14
- Direct access to the founder
What we ask
- One agent decision workflow to connect (staging or test is fine)
- An 8 to 12 week pilot
- A 30-minute check-in every two weeks
- A named technical contact
- Optionally, an anonymised case study at the end
CARC already runs against a hospital triage agent and a loan review agent. Watch the demo ↗
Apply
Two minutes. Tell us about the decision your agent makes or supports.
30-minute fit call
We walk through your workflow and agree whether a pilot makes sense.
Connect one agent
Connect one agent workflow to CARC and start the pilot.
08. Platform Principles
Runtime oversight for regulated, high-stakes AI.
CARC helps produce the records needed for human oversight and audit when oversight, intervention, auditability, and post-market monitoring move from policy language into deployed product behaviour.
Article 14
Effective Human Oversight
Human oversight evidence must show more than a named reviewer. CARC's Decision Controller defines when a human reviewer can intervene, what they saw, how the agent behaved, and whether oversight was operational at the point of use.
Annex III
High-Risk Use Cases
Annex III lists AI used in employment, creditworthiness assessment, and life and health insurance risk assessment and pricing as high-risk. CARC helps produce the records needed for human oversight and audit of agents operating in these areas.
GSPR 14.2(d)
Intervention and Override Evidence
In healthcare, MDR AI software needs clear evidence of user intervention, escalation, and override paths where clinical decisions are influenced by automated processes. CARC's Evidence Engine helps make those paths explicit and reconstructable.
Surveillance
Behaviour Monitoring After Deployment
Post-market AI monitoring depends on knowing how the agent behaves across real cases: recommendations, confidence shifts, escalations, deferrals, overrides, reviewer actions, and repeated-case consistency. CARC's Runtime Metrics engine is built for exactly this.
09. Company
Giggle AI Innovation builds runtime infrastructure for AI systems making consequential decisions. Our flagship product, CARC, is the runtime control layer and audit trail for AI agents in high-stakes decisions.
I work directly with design partners integrating CARC into their AI systems. If your product is moving into deployment, audit, or post-market monitoring, request a demo.
Thokozile Phiri, Founder, Giggle AI InnovationQuestions
What is CARC?
CARC (Consequential Agent Runtime Control) is a runtime control layer for AI agents making consequential decisions. It sits between your AI and the workflow it acts in, making agent behaviour observable, governable, interruptible, and reconstructable in production.
Is CARC live today?
CARC's Developer Platform, API, and Control Console are implemented and available to design partners under private preview. Public hosted access is rolling out. Request a demo to get access now.
Does CARC replace our existing AI?
No. CARC does not replace your AI models or agents. It adds a runtime control layer alongside what you already have. Keep your AI, add runtime control.
Is this a SaaS platform or an advisory engagement?
CARC is a software platform: a Runtime Engine, Decision Controller, Evidence Engine, Policy Engine, Risk Engine, and Control Console with a Developer SDK and API. Giggle AI Innovation also works directly with design partners during private preview to integrate CARC into their systems.
Who is CARC for?
Teams deploying AI agents in high-stakes decisions such as hiring, lending, insurance, and healthcare: CTOs, Heads of AI, product and platform engineering teams, and risk, regulatory, or quality leaders preparing AI agents for deployment, audit, or post-market monitoring.
How do I become a design partner?
Apply on the design partner page. It takes about two minutes. If it looks like a fit, we book a 30-minute call, then connect one of your agent decision workflows to CARC for an 8 to 12 week pilot, free of charge.
Request Demo
Know how your
AI agent behaves
after deployment.
Request a demo to see how CARC governs, reviews, and reconstructs agent behaviour in production, and how to join as a design partner today.
No preparation required. No commitment beyond the conversation.