Fragmented evidence
Voice, SMS, email and tool calls each leave a log in a different system. Nobody can reconstruct one session end to end, so nobody can defend it.
Founder & partner programme · Canada
We built the trust layer for AI agents. Build the next layer with us.
Netwit runs agents inside Fortune 50 operations and built the control plane that proves what they did. Assure registers every agent, enforces policy in under 25 ms, and keeps a hash-chained record in Canada. If you build agents, models or vertical tools, this is the runtime, the compute and the distribution you would otherwise spend two years assembling.
flight recorder · live ca-central
0x8f2a agent.register(assure) signed
0x3c71 policy.evaluate(consent, dnc) 2.1 ms
0x9b04 tool.call(crm.write) allowed
0x5d18 recorder.commit(sha256) chained
0x7e42 evidence.pack(law25, pipeda) sealed
Illustrative trace. No customer data.
14M+
calls a year
voice agents, one Fortune 50 deployment
12
agents on one runtime
MCP-native, audit trail by default
3 → 6
DGX Spark systems
in our own British Columbia lab
5 weeks
to a go / no-go
six gates, either side can stop
The accountability deficit is the buying criterion.
Every platform now ships a console that watches its own agents. Almost nobody can prove, across vendors, what an agent did, on whose authority, with which data, and whether the policy held. When an agent leaks data or breaches consent rules, the deployer carries the liability — and the deployer is the one who cannot produce the file.
Governance that stops at the vendor
A platform console governs its own agents. A CIO running four vendors needs one registry and one ledger that outlives any of them.
Insurers cannot price what they cannot see
Underwriters ask for control evidence, not assurances. Without a standard pack, agent exposure lands in silent cover, or in a decline.
Not a roadmap. A track record with a control plane on top.
Every control in Assure came from operating agents for paying clients — voice, messaging, email, clinical — before it was a product. The modules below are the ones we will demonstrate live, with their status stated exactly as it stands.
Agent Registry
Cryptographic agent identity, client attribution, version lineage, scoped roles and whitelisted tool endpoints.
Built Runtime Policy Engine
Sub-25 ms inline checks: AI disclosure, consent verification, do-not-call lookups, opt-outs and a per-agent kill switch.
Built Flight Recorder
SHA-256 hash-chained record of prompts, outputs, tool calls and policy decisions, in Canada, with per-tenant keys.
Built Evidence Packs
Dossiers mapped to Law 25, PIPEDA, CASL, ISO/IEC 42001 and AIUC-1, with draft privacy impact assessments.
Built Reference Connectors
Native on LangGraph and LiteLLM; commercial connectors for voice platforms, n8n and custom MCP endpoints.
Extending Evaluation Lab
Scheduled red-teaming, prompt-injection stress tests and adversarial evaluation on DGX Spark, signed into the same ledger.
Extending Four layers. Plug in wherever you are strongest.
This is not a single product with an API bolted on. It is a lab, a runtime, a trust plane and a portfolio, and each layer is a legitimate place to attach.
Lab
DGX SparkSovereign podsEval corpus
Runtime
12 agentsMCP toolsPer-agent RBAC
Trust plane
RegistryPolicyRecorderEvidence
Solutions
VoiceMessagingEmailClinical
Three ways to plug in — each with a first trace in days.
You do not have to adopt the whole stack. Pick the seam that matches your product and prove it inside a sprint.
Gateway connector
Route your agent's calls through the Assure gateway for registry, policy checks and recording. The reference SDK is live; a first trace typically takes days.
MCP tool endpoint
Expose your product's capabilities as MCP tools so any governed agent can call them with scoped permissions and a full audit trail.
Signed evidence adapter
Send test, observability or identity results into the ledger through a cryptographically signed import bridge. One adapter, one four-week sprint.
What production already proved.
Customers are under NDA, so we show the archetype and the number. Each one is verifiable in the data room, and none of them is a projection.
$150M
annual handle-cost savings
Fortune 50 telecom voice
47,000
claims a week
Fortune 50 insurer, built software
2.3M
bank documents a year
91.4% first-pass accuracy
90%
less compliance workload
Fortune 50 security operations
2,400
articles a week
Swiss national newspaper
99.99%
managed email SLA
PostMTA Cloud, 16 regions
Anonymised outcomes, verified individually under NDA. No composite score, no review stars, no customer counts.
Eight things the market is racing toward — already being built.
Each sits on top of what already exists, so the build is quarters, not years. Each has an open seat, and the partner shapes the product rather than reselling it.
Q1 2027
Agent Identity & Access Vault
Non-human identity is a ~US$1B security category.
Short-lived scoped credentials, per-agent revocation, credential-use ledger.
Q1–Q2 2027
A2A + MCP Trust Gateway
Agent-to-agent calls need policy at the boundary.
Signed delegation, cross-organisation logs.
Q2 2027
Agent Commerce Guard
Payment rails for agents are live and unguarded.
Spend limits, consent, receipts, dispute evidence.
Core built
Voice Assure
EU transparency duties and consent enforcement on live calls.
Bilingual disclosure, DNC enforcement, automated QA scoring.
Q2 2027
Governed Agentic SOC
Security operations are being rebuilt around agents.
Scoped tools, human approval gates, tamper-evident case records.
Q4 2026–Q1 2027
Clinical Assurance Kit
Provincial AI-scribe guidance demands consent and deletion.
Consent capture, retention enforcement, review logs, PIA drafts.
Q1–Q2 2027
Physical AI Proving Ground
Simulation and robotics converge; failure testing is costly.
Physics-grounded gyms, synthetic failures, actuator kill switches.
2027
Sovereign Inference Gateway
Canadian procurement now asks where inference runs.
Policy-driven gateway and evidence in front of Canadian pods.
Dhiraj Chatpar · Founder
Eighteen years building AI and infrastructure across Canada, the UAE and India. ISC2/CISSP-credentialed, fractional CTO, and founder of the Lower Mainland Business Alliance. Operates the runtime and the lab this programme runs on.
Five weeks to a decision, not five months to a maybe.
One call, a scored fit, then a thirty-day sprint on our infrastructure. Either side can stop at any gate.