The runtime behind every CLAiRE appliance.
CLAiRE is not a chatbot. It is a platform — an AI runtime that orchestrates workflows, reasons over your knowledge, and double-checks itself before any answer reaches a human. The same engine powers every CLAiRE appliance, so each one inherits a decade of governance work on day one.
Eight things that turn a chatbot into a platform.
Every appliance — MAiGRATE, Polarion Copilot, ATR, DRIFT, APQR, and the three more on the way — runs on the same eight capabilities. Add a new appliance and it inherits all of them on day one.
One workspace, one IA
Overview, chat, workflow builder, knowledge atlas, evaluation lab, and governance — every customer, every appliance uses the same interface.
A conductor for your processes
The engine that runs every workflow. It knows what step comes next, what to do when something fails, and how to pause for human approval.
AI grounded in your vocabulary
One AI brain trained on your controls, deviations, and products. It cannot make up a term you have never defined — every query is pre-checked against the real shape of your data.
Double-checked before a human sees it
Each AI answer is scored on a confidence band. Green proceeds. Amber routes to a person. Red is blocked. A library of 22 lessons-learned catches the patterns we have seen break agents before.
Safe by the time it reaches the AI
Personal data is replaced with safe tokens at the boundary. Four layers of guardrails check what goes in, what the AI can use, what tools return, and what comes back out.
Four kinds of memory
Permanent knowledge of your business. Memory of past decisions. A scratchpad for the current task. And a versioned library of how the platform does things. Together: cognition, not chat.
A safety net around the AI
A semantic layer wraps every AI emission — checking before, reasoning after, and tracking hypotheses while the agent thinks. It does not intercept; it stands ready to catch.
A platform that watches itself
Anomaly detection, model monitoring, automatic remediation, and SLA escalation. The platform learns from its own failures and adjusts before customers notice.
Our semantic layer is a sidecar, never a bottleneck.
Most platforms force every AI call through a triplestore. CLAiRE does the opposite — the semantic layer sits beside the runtime, checking before, reasoning after, and never standing in the way.
It knows your business
A registry of your terms — controls, deviations, CAPAs, products. The platform reasons over your reality, not generic knowledge.
It checks before it acts
Before any AI-generated query touches your data, the platform confirms it references things that actually exist. Hallucinated relationships are rejected outright.
It derives new facts after the fact
When records change, the platform applies your rules to find what changed downstream — gaps, inconsistencies, controls that need re-running.
It keeps notes while it works
During a task, the AI maintains a scratchpad for hypotheses and partial findings. The notepad is discarded when the task ends unless you keep what it found.
From your question to a defensible answer.
Every request — chat, workflow, or API call — moves through the same gates. Nothing reaches a human until each one has signed off.
You ask
Chat, build a workflow, or hit the API.
The question is grounded
Pre-validated against your vocabulary — anything hallucinated is rejected.
The right agent runs
Personal data tokenised; tools called; LLM reasons in your domain.
The answer is scored
Rules + an AI judge double-check the output. Green proceeds; amber escalates; red blocks.
Evidence is stamped
Every action signs the previous one. The whole run is replayable for the next audit.
Three commitments, built into every product.
You can see what the AI did
Every answer comes with a “Why” panel and a replayable record. Open any decision and walk it back to its source. Drift, anomalies, and degraded models surface on a dashboard — not in your next inspection.
The platform heals itself
Failed calls retry. Bad models swap to a backup. Late approvals escalate. Continuous compliance scans run in the background. Every agent has its own library of golden test cases, run on every save.
Built for the auditor
Six roles, granted at the record level. Personal data tokenised before it ever reaches an AI model. Five-tier classification with attribute-based access. A tamper-evident evidence chain on every action. SOC 2, ISO 27001, and ISO 42001 ready.
Boring choices. Reliable foundations.
The substrate is intentionally unsurprising. Proven databases, proven message buses, proven providers. Excitement belongs in the application — never the foundation.
OpenAI, Azure OpenAI, and Anthropic — routed per tenant and per agent, with automatic failover.
SQL Server, Neo4j, Redis, and Azure Blob — the boring choices, made for reliability.
Your business knowledge lives as a connected graph, not a pile of documents.
On-prem deployments can use a locally hosted model — your data never leaves your network.
Use the search backend you already have — Postgres, Azure AI Search, or Chroma.
Every decision the platform makes can be streamed out as an event for downstream systems.
Built by the people behind FDA CSA.
Compliance Group co-founded the FDA-Industry Computer Software Assurance team and contributed to the FDA CSA guidance. Twenty-five years inside audit rooms — now compiled into CLAiRE.