From the OBEL™ Intelligence Brief · April 2026
Enterprise Architecture
It's not a capability problem. Frontier models are extraordinarily capable. The barrier is governance — and it compounds with every agentic deployment. Every organisation with a compliance function faces it. Most have no answer.
Agentic Security
An autonomous agent may execute dozens of LLM sub-steps, access file systems, query databases, and call external APIs — all without human review. Each tool call is a potential data egress event, a sovereignty violation, and an audit gap.
Autonomic Discovery
Any fixed pattern library is a snapshot of yesterday's threat landscape. OBEL™'s Autonomic Discovery Engine closes this gap continuously — the security perimeter grows with each organisation's interaction patterns, not when a vendor ships a patch.
Works with the providers your team already uses
Anthropic
Claude
OpenAI
GPT · o-series
Gemini
Meta
Llama via Groq
Azure OpenAI
Microsoft
Groq
LPU Inference
And more
View integrations
Live demo
Type any message and see how OBEL classifies it before it reaches any AI model.
Gateway Architecture
Scrubbing and classification run at $0.00. Inference only starts if the gate passes. A blocked request costs nothing — the model never sees it.
Receive
Raw prompt enters the gateway. OBEL intercepts before any AI model sees the text.
Scrub
PII, API keys, and injection patterns stripped. The cleaned text advances — the original never leaves.
Classify & Gate
ARGUS-i™ applies sovereign schema. PROTECTED+ content is hard-blocked here — the model is never called.
Inference
Only gate-cleared, scrubbed prompts reach the model. You're billed on the cleaned input — never the raw text.
Audit
Every interaction committed to the immutable vault before this step completes. No audit write, no inference proceeds.
Other AI gateway providers may offer PII scrubbing — but only on paid tiers, without sovereign classification and without a pre-inference audit write. OBEL™ runs this pipeline on every request, on every plan, with no bypass.
Platform capabilities
Document Studio
Draft contracts, reports, code, and diagrams with frontier AI — then export, audit, and reuse them. Every generation request passes through ARGUS-i™ before a word is drafted.
Template Intelligence
Every org gets a pre-built library of NDAs, employment contracts, service agreements, statements of work, consulting agreements, and more — each with required fields, formatting rules, and placeholder validation. Define your own types for custom workflows.
Secure AI Generation
ARGUS-i™ runs on every generation request — classifying intent, scrubbing PII, and blocking sovereign violations before a word is drafted. Documents, code, diagrams, reports, and workflow outputs are all treated as governed artifacts, not free text.
Artifact Vault & Export
Generated artifacts are stored in your org vault — version-controlled and tied to the audit session that produced them. Export any document as a branded PDF. Every export is attributable to a user, a session, and a complete governance record.
Agentic Foundry
Deploy autonomous agents against Notion, Jira, Salesforce, Slack, and 10+ other enterprise apps. Mention @AgentName in chat to dispatch a task instantly. Every tool call is inspected by ARGUS-i™ and can require a human sign-off before it executes.
@ Mention Dispatch
Type @AgentName followed by a task in any conversation and OBEL™ dispatches it to the correct agent — no context switch, no separate interface. The agent's tool calls, reasoning steps, and final response stream back inline.
HITL Governance
Configure agents at any authority level. At Level 1, agents observe and report. At Level 2, agents can act — but high-risk tool calls pause the run, surface a proposed action to an administrator, and wait for approval or rejection before proceeding. The SSE stream stays live throughout.
Blueprint Builder
The Blueprint builder lets you configure an agent's identity, connected apps, and read/write permissions in a single view. Authority level is automatically derived from the permissions you grant — read-only agents can never write, regardless of what an LLM requests.
ARGUS-i™ Detection Packs
ARGUS-i™ ships with certified detection packs covering every major regulatory jurisdiction. Each pack is a portable bundle of regex rules, replacement tokens, and severity classifications — applied to every message before it reaches inference. And if your jurisdiction isn't covered yet, you can add it.
Global
Universal patterns
Five Eyes
FVEY intelligence frameworks
European Union
GDPR · ePrivacy Directive
MENA
UAE · KSA · Egypt
Asia-Pacific
SGP PDPA · AUS Privacy Act
Community
Submit for your jurisdiction
Rule Anatomy
Each detection rule carries a pattern, a jurisdiction-scoped replacement token, a severity level (low / medium / high / critical), and a PII event type. ARGUS-i™ evaluates every active rule against every message — before a single token reaches the model.
Built-in Coverage
Global covers universal patterns (emails, phone numbers, API keys). Regional packs add jurisdiction-specific IDs: UK National Insurance, EU VAT numbers, Australian TFNs, UAE Emirates IDs, Singapore NRICs, and more. Every built-in rule is reviewed and certified by ninthLABS.
Community Packs
Write a pack.json, validate it against the open schema, and submit it through the OBEL™ governance panel. ninthLABS reviews each pack for schema validity and safety before approving it — once live, it's available to every organisation on the platform.
Retrieval-Augmented Generation
Standard RAG retrieves documents and hands them straight to the model. OBEL™ runs every retrieved chunk through ARGUS-i™ classification and PII scrubbing before a single token reaches inference — at retrieval time, not just at upload.
Double Scrub
Most RAG pipelines scrub once at upload and trust the result forever. OBEL™ re-runs the PII scrubber and ARGUS-i™ classifier on every chunk at the moment of retrieval, so updated classification rules apply retroactively — without re-indexing.
Source Citations
OBEL™ surfaces which chunks grounded each response so your team knows exactly where an answer came from. Every retrieval event is written to the immutable security event log — auditable, exportable, and SIEM-ready.
Zero Training Risk
Chunks are stored in your own Supabase instance under pgvector, never sent to a shared embedding service, and never used to fine-tune an external model. Your organisation's row-level isolation boundary holds — even during retrieval.
Start with a 14-day free trial — full Individual Pro access, no credit card required. Upgrade to a team plan when you need multi-member workspaces and full governance.