What does Bitligence build?

Intelligent agents for consequential enterprise work, plus the runtime governance that makes them safe to act. The intelligence comes from foundation models; the governance runtime is the product Bitligence builds.

Why not just use a foundation model and an agent framework?

Those supply intelligence. They do not decide, in the path, whether a specific action is authorized given the agent's capability profile, the active policy, the live system state, and the reversibility of the action. That decision — signed and replayable — is the runtime governance Bitligence builds.

Intelligent agent systems · governed at runtime

Build intelligent agents you can trust to act.

Q: Which industries does Bitligence serve first?

Insurance (P&C claims) is the live beachhead: Atlas handles FNOL voice intake and Praxis runs the claim file, with every decision scored and sent to an oversight inbox. The same governed runtime extends to financial services, healthcare, government, legal, and enterprise operations.

Agents that observe, reason, and take real action in consequential work. Every Agentic action passes a governance gate before it executes — policy-checked, capability-verified, logged to a signed ledger.

Intelligence: From the foundation model — commodity, improving for free as the labs ship.
Trust to act: The runtime governance we build — every action checked in the path before it executes.

Book a demo ›How the runtime works

Built on our own platform — Keelson to build, AO to govern.

Live product · Assurance Observatory — governance overview

Assurance ObservatoryICW Group / claim-workspaceSearch ⌘KLLena ADMIN

Assurance Observatory runtime · governance overview

Good morning, Lena.

Posture · last 24 hours1 critical3 major8 minor

12 events · 1 critical needs attention

24h agonow

⚡ Live throughputlive runtime ›

47fires / min▲ +7/min vs 60m

Routing mix · last 24h

78% auto18% HITL4% blocked

Compliance coveragebuild report ›

2/ 4 regulators fully covered

NAIC AI Model Bulletin 202311/12 92%

Colorado SB21-169 20218/9 89%

NYDFS Circular Letter No. 1 20197/7 100%

State DOI baseline 202414/14 100%

Agents in runtime

Top-level2

LLM3

Active policies

1010 active

AI Ops approvals

3pending · oldest 18h

Coordination changes2Policy promotions1

agent connected·guidewire claimcenter·last run 12 min ago·14 actions overnight

≤10 mscritical-path budget

fail-closeddefault on timeout

signedevery decision, replayable

policy = codeversioned, attested

The 2026 reality

A working agent demo takes an afternoon. Production takes the runtime.

What the demo leaves out is exactly what every regulated buyer’s review asks for. That gap — not the model — is why most agent prototypes never ship. It is the gap we close.

An afternoon

The prototype

A foundation model
A set of tools
A nice-looking UI
A happy-path demo

The gap

Everything regulators ask for lives here — not in the model.

What production needs

Attestation chain & signed audit
Rollback path & override audit
Multi-tenant isolation & inbox SLA
Rule-promotion trail
Bidirectional system-of-record write that survives partial failure

Runtime governance — the product

Every agent action passes one synchronous gate before it executes.

Not a model card, not a dashboard you read after the incident. A decision made in the path — permit, transform, escalate, or deny — in about 10 ms, fail-closed by default, and written to a signed, replayable ledger.

The request path — every action stops at one gate and is decided before it executes: permit, transform, or deny

Before the action

Four checks. One path. Zero unauthorized moves.

Every agent action is evaluated across four dimensions. If it fails anywhere, the action stops.

Capability

What outcomes this agent is authorized to produce — not a tool list. A multi-tool chain that produces an unauthorized outcome is still a violation.

VerifiedOutcome within authorized scope

Policy

Which business, security, compliance, and budget constraints apply to this action, right now.

VerifiedPolicy allows this action

Context

What is known at this point in the workflow — evidence, priors, the agent's read of the task.

VerifiedContext supports the action

State

What is happening across shared systems before the action fires — close-freezes, locks, in-flight writes.

VerifiedSystem state is safe

StopIf any check fails, the action is blocked.

What makes it different

Three primitives most agent platforms don’t operate on.

Plenty of products claim a “trust layer.” What matters is what the platform actually reasons over — three primitives an examiner can interrogate, and most agent stacks don’t have.

Primitive 01

Action classes, not tool lists.

Every privileged operation declares an action class — mutates-state, sends-external, irreversible — and an aggregation rule. Detectors reason about aggregate effect across the call graph, not per agent, per tool.

Sample registered classes · live

modify-reservemutates-state318
send-externalsends-externalreveals-info1,847
approve-terminalmutates-stateirreversible42

Primitive 02

Capability profiles, not RBAC.

Each agent declares the outcomes it's authorized to produce — not the tools it may call. The runtime checks whether the aggregated effect of a tool chain fits the profile, so composition violations get caught.

Composition verdict · simulator

profile · file-handler

+ 3 tools selected

mutates-statesends-external

Verdict · Excluded breach

Excluded outcome: send-external

Primitive 03

Session graphs, not event logs.

Every event stamps causal_parent_id and delegated_from_id, so the audit trail is a queryable directed graph, not a flat log. Structuring detection, conflict detection, and capability composition all need this.

Synthesized session graph

each edge stamped with confidence delta + causal parent

Horizontal agents

One governed agent, deployed across every industry.

A vertical build goes deep in a single domain. A horizontal agent runs the same job across all of them — and the same runtime governs every action it takes.

The CX Agent is the first, live today.

Insurance

Claims, policy, billing support

Healthcare

Member support, prior auth, benefits

Fintech

Account support, payments, disputes

Pharma

HCP support, patient programs, safety

Legal

Case intake, status, documents, support

Runtime GovernanceEvery action checked in the path, signed, and replayable.

Policy EnforcementAudit TrailAttestation ChainEvidence Retention

One Agent. Every industry.Consistent experience. Domain-aware outcomes.

One governed experienceOne runtime across allDomain context · Shared intelligenceLower cost · Faster scale

Meet the CX Agent ›Or explore by industry

Verticals

The same governed runtime, proven on consequential work.

Insurance is live today. The governance layer is domain-general — action classes and capability profiles don’t assume your workflow — so the same runtime extends across regulated operations.

5Verticals

1Unified runtime

200+Action classes

100%Audited

Insurance

P&C claims — FNOL voice intake (Atlas) and claim-file handling (Praxis), every decision scored and sent to an oversight inbox. The workflow carrying production load today.

Platforms

GuidewireDuck CreekACORD

Explore →

Pharma & life sciences

Pharmacovigilance & safety operations — intake and triage adverse-event reports, draft the case narrative, route signals for medical review, every step attributable for audit.

Platforms

MedDRAE2B (R3)Veeva Vault

Healthcare payers

Prior authorization & coverage review — structure the clinical request against medical-necessity criteria, attach evidence, and route exceptions to clinical review.

Platforms

HL7FHIRX12 278

Fintech & lending

Origination, KYC/AML, and fraud decisioning — document intake, identity and sanctions screening, credit calls. Routine approvals on earned rules; edge cases to an underwriter.

Platforms

PlaidEquifaxOFAC

Legal & compliance

Contract & obligation review — clause extraction, deviation detection, redline proposal. A hard capability boundary between summarizing and binding modification.

Platforms

iManageNetDocuments

Multi-agent coordination

Guarding a single agent is the known part. The risk lives between them.

Put several autonomous agents together and they contend for the same resource — racing, duplicating, deadlocking. Bitligence governs the coordination between them: every conflict is resolved before any agent acts, so one proceeds and the others wait, reroute, or are stopped.

Watch the governor arbitrate: agents post intent to a shared state; one proceeds, the others wait, reroute, or are denied. Select a deck to see it across domains.

Coordination deckFinancial close01 / 06live

Beyond coordination

One control point, many problems.

Once every agent action routes through one governor in the path, that single control point addresses problems most teams would otherwise solve with five separate tools — security, reliability, audit, anti-collusion, and cost.

Coordination · Live

Arbitration in the path

Agents contend for the same resource; each conflict is resolved before any of them acts — one proceeds, the others reroute, wait, or are stopped.

in-path · ~8 ms · one proceeds, others wait / reroute / deny

How autonomy is earned

Day 1: every agent decision is reviewed. Day 90: only the ones that haven’t earned a rule.

When agreement on a category crosses ≥95% over ≥20 samples, the runtime surfaces the rule with its supporting evidence. Your operator approves it once. That category stops landing in the inbox; the agent runs it under the approved rule, with the attestation chain preserved.

Inbox volume · % of decisionsRules active · cumulative

Inbox volume vs. rules active90-day window

Synthetic from observed deployments. Real curves vary with workflow complexity and operator response rate.

1Day 1

Everything to the inbox.

100% of agent decisions surface to the inbox. Full human review. No autonomy assumed.

Inbox volume ~100%

2~30 days

Rules are earned.

High-agreement categories generate rule candidates. The operator approves once per category.

Rules active · growing

3~90 days

Routine runs on autopilot.

Routine cases close automatically under approved rules. The inbox holds only novel and high-stakes work.

Inbox volume ~10% or less

What ships today

The surfaces your team operates every day.

The exception feed, the policy catalog, the proposal queue, and examiner-ready reports — each running in production today, operated by the people who run governance day to day.

Operator surfaces · live in production01 / 04

Workflow baselines

Learned distributions, not generic thresholds.

Per task class, the platform learns observed depth, fan-out, and retry density — then compares new graphs against the customer's own history, not a vendor-set rule.

depth · nodes

Security & compliance

Runs in your boundary, audited end to end.

Deploy in your own boundary. Every decision is signed and replayable for audit.

AttestedSOC 2 Type IIISO 27001HIPAAFedRAMP-readyEU AI Act

01 · Deployment

Your boundary

On-prem, VPC, or fully air-gapped. Your data never leaves your perimeter.

Your environment

On-premVPCAir-gapped

02 · Evidence

Signed & replayable

Every permit / redact / deny is cryptographically signed and reconstructable.

PermitRedactDeny

Signed · immutable · replayable9f7c2a1e8b6d…a3b771a9b2d5e

03 · Trust center

Docs & reports

Architecture, sub-processors, and current attestations. Request access →

Security architecture

Sub-processors

AttestationsUp to date

Put a trustworthy agent on your workflow.

Start where autonomy matters, the actions are valuable, and control can’t be bolted on later. We’ll scope it with you.

Book a demo ›See it in insurance