The compliance-grade orchestration layer for production AI agents
Every regulated team is rebuilding the same AI stack — connectors, durable execution, model routing, compliance, audit trails. Magnitude Engine is that plumbing, productized. Turn an administrative process into an auditable, infinitely scalable workflow in hours, not quarters.
Zero to production in 30 minutes. Mature the prompts from there.
# A workflow is a graph. Each step: input → prompt → output.
workflow: claims-triage
region: eu-west-1 # data-residency boundary
compliance: [GDPR, ISO27001]
steps:
- id: intake
connector: claims_system # pull from internal systems
model: claude-opus-4-8
prompt: "Extract claim entities and flag missing docs."
- id: triage
model: claude-opus-4-8
prompt: "Classify severity. Route fast-track vs. review."
on_low_confidence: escalate_to_human
- id: payout-check
connector: policy.db
prompt: "Validate payout against policy terms."from magnitude import Workflow, step
claims = Workflow("claims-triage", region="eu-west-1",
compliance=["GDPR", "ISO27001"])
@step(model="claude-opus-4-8", connector="claims_system")
def intake(claim):
return "Extract claim entities and flag missing documents."
@step(model="claude-opus-4-8", on_low_confidence="human")
def triage(claim):
return "Classify severity. Route fast-track vs. review."
# Durable execution + infinite scale on AWS Step Functions.
run = claims.deploy().invoke(claim_id=10482)
print(run.trajectory_url) # trace every step, every prompt# Or build it visually — spreadsheet/canvas UI for business users.
┌──────────────┐ ┌──────────────┐ ┌──────────────┐
│ INTAKE │ ──▶ │ TRIAGE │ ──▶ │ PAYOUT CHECK │
│ claims_sys │ │ severity │ │ policy.db │
│ claude-opus │ │ + routing │ │ validate │
└──────────────┘ └──────┬───────┘ └──────────────┘
│ low confidence
▼
┌──────────────┐
│ ESCALATE │ → Slack / Workspace
│ to human │
└──────────────┘
Each node = input (connector) → system prompt → output.
Pick the model and AWS region per step. Ship.Every regulated industry is independently rebuilding the same AI agent stack.
Insurance claims teams, financial-research desks, pharma compliance groups, and back-office HR functions are all solving identical infrastructure problems in their own silos — each hiring expensive AI engineers to reinvent the plumbing before they ever ship a feature. That fragmentation is the opportunity.
AI isn't a research problem anymore. It's an integration, reliability, and compliance problem.
The hard part was never calling a model — it's everything around the model. Every team rediscovers this the painful way, in production.
Bespoke, brittle integrations
Every team writes its own connectors to its own corporate systems — then maintains them forever.
Reliability is hard
Multi-step agents need durable execution, retries, and state management. Most teams learn this in production.
Compliance is a blocker
In insurance, finance, pharma and healthcare, nothing ships without GDPR, ISO, data-residency and audit. Bolting it on later is slow and expensive.
Agents are black boxes
When an agent gets it wrong, teams can't trace why — which step, which prompt, which input led there.
Talent is scarce
The reflexive answer is "hire AI engineers" — months of ramp and salary rebuilding infrastructure that isn't differentiating.
Two quarters of plumbing
A long tail of vertical AI teams each spend their first two quarters rebuilding the same substrate before delivering any value.
A workflow is a graph. Each step has three parts.
Chain the steps and you have a complete process — claim triage, a compliance check, a research brief — running reliably in the cloud.
Pull from your systems
Data comes in through connectors the business defines — in code or via the SDK — into your corporate systems and document stores.
Inject your model of choice
A system prompt runs against the model you pick — Claude, Llama, Gemini, OpenAI — including models served through Amazon Bedrock.
Pass to the next step
Each output feeds the next node. Validate, branch, or escalate to a human — all with durable execution underneath.
Author the same workflow three ways — visually in a canvas UI for business users, declaratively as YAML for complex DAGs, or programmatically via the SDK (Python, Go, and more) with first-class API keys.
Not one axis — the combination that's hard to replicate
Durable execution is necessary but not sufficient. Magnitude is durable and compliant and auditable and domain-aware.
Durable + scalable by default
Built on AWS Step Functions, every workflow gets durable execution and effectively infinite scale out of the box. You pick the model and the AWS region — so the same workflow runs inside the compliance and data-residency boundary your industry requires.
- ✓Retries, state & recoveryLong-running, multi-step DAGs survive failures without you writing the orchestration.
- ✓Model- and region-flexibleRoute to Claude, OpenAI, Gemini, Llama or Bedrock per step. Run in the region you must.
- ✓On-prem path on the roadmapRun the engine on your own GPUs, orchestrated via Kubernetes, entirely inside your walls.
| Capability | Build it yourself | Durable-execution frameworks | Magnitude Engine |
|---|---|---|---|
| Durable, scalable execution | Months to build | Yes | Yes — day one |
| Compliance certifications | DIY, slow | Not included | GDPR, ISO — in place |
| Deep corporate integrations | Bespoke per team | Generic | Productized connectors |
| LLM-native observability | Rare | Logs, not trajectories | Trajectory-level |
| Cross-domain workflow learnings | Siloed | None | Baked in |
| Time to compliant production | Two quarters | Weeks–months | Hours |
Trust, certified up front
A customer with an urgent need ships faster because the platform already carries the data certifications they'd otherwise spend months acquiring.
The wedge into regulated industries
This is the thing competitors treat as an afterthought — and exactly where the compliance bar is highest.
- ✓Auditable by designEvery run is traceable and explainable, end to end.
- ✓Trajectory-level observabilityDive into the exact reasoning path an agent took, step by step, to debug why an outcome happened.
- ✓Where your team already worksDeep Slack & Google Workspace integration — a Magnitude bot sits alongside employees. Hand it a run ID and it walks you through what the agent thought and did.
High-volume, rules-heavy, compliance-sensitive work
Any industry with structured back-office and administrative processes — starting where the pain and the compliance bar are highest.
Insurance
Claims intake, triage, and processing — high volume, high compliance bar, no incumbent owns the engine beneath it.
Financial services
Agentic research analysts and reporting that run reliably and leave an audit trail.
Pharma & healthcare
Fast-track compliance and regulated administrative workflows inside the data boundary.
Real estate
Document- and process-heavy operations, automated end to end.
HR & recruiting
Repeatable administrative processes across every company's back office.
Vertical AI builders
Stand on Magnitude instead of rebuilding the substrate — the infrastructure layer for regulated agents.
Well-funded teams are already building this — one vertical at a time
Each rebuilds the full stack inside a single domain. Magnitude generalizes the engine underneath them.
AI for financial-advice firms
Well-funded players have built their own data capture, workflow automation, compliance suites, and deep two-way back-office integrations — serving hundreds of firms and thousands of advisers.
Takeaway: a single vertical justified an entire purpose-built stack — connectors, workflows and compliance all rebuilt in-house.
AI teammates for healthcare admin
Referral triage, waiting-list validation, booking, and patient comms — architectures almost identical to Magnitude's per-step model, but hardwired to one domain.
Takeaway: the workflow-graph-with-validation pattern works and sells — the engine underneath should be horizontal.
Tech-enabled claims services
Providers that run claims as a service — human talent plus proprietary tech, fraud detection, and policy-specific workflows.
Takeaway: many regulated buyers want the outcome done for them — a natural design partner or customer for Magnitude, not a rival.
Agentic platforms in one modality
The closest structural analogues pair a no-code builder, a developer SDK, enterprise compliance by default, and proprietary models — but stay inside a single modality, such as conversational voice for contact centers.
Takeaway: the builder + SDK + compliance-by-default shape scales and raises money. Different modality — Magnitude orchestrates back-office processes, not real-time conversation.
vs. the vertical players, Magnitude is horizontal and multi-vertical by design — model- and region-flexible, with an on-prem path. The substrate beneath the next wave of vertical agents, not another sibling competing in one lane.
Buildable today in a way it wasn't 18 months ago
Frontier models are finally good enough to run real administrative work — but the surrounding infrastructure hasn't been productized. Whoever ships the trusted substrate captures the platform layer.
Cloud today. Platform next. Run-anywhere after.
Win one beachhead vertical deeply, then generalize the engine underneath it.
Cloud, model-flexible, compliant
Visual + YAML + SDK authoring, durable execution on AWS Step Functions, model and region selection across Bedrock, core compliance certifications, and trajectory observability.
Platform & ecosystem
Public SDK in Python, Go and more with API keys — so vertical builders stand on Magnitude instead of rebuilding infra. Deeper corporate integrations and the Slack/Workspace debugging bot.
Run anywhere, including on-prem
Host the engine on customers' own GPUs on-premise, orchestrated via Kubernetes — so the most tightly regulated environments run AI workloads entirely inside their own walls.
A usage-based platform on top of model costs
Built to scale with the work it runs — not to gate teams behind seats.
Consumption
Per-workflow / per-execution pricing on top of model costs — you pay for the processes you actually run.
Enterprise
On-prem deployment, advanced compliance, and dedicated support for the most regulated buyers.
Design partners
Hands-on onboarding for early insurance and finance partners shaping the roadmap with us.
Request early access
We're onboarding a small set of design partners in insurance and financial services. Tell us about your workflow and we'll be in touch.
No spam. We'll only reach out about your early-access request. Investor? Read the vision memo →