How It WorksWhy SphereIQKnowledge HubGet Started
SphereIQ Documentation

Everything You Need to Deploy AI Teammates

Platform overview, installation guides, configuration reference, and security policies — all in one place.

OverviewInstallationConfigurationSecurity & Privacy

Platform Overview

SphereIQ deploys autonomous AI teammates into operational workflows across four regulated industries.

Unlike generic chatbots, AI teammates are trained on industry-specific processes, connected to core enterprise systems, and governed by audit-ready controls. Each teammate handles a defined function — claims intake, prior authorization, loan processing, quality inspection — and operates 24/7 with human oversight for high-stakes decisions.

How it works

1

Receive

Inbound work arrives via API, email, portal, or system event.

2

Retrieve

The teammate searches internal knowledge bases and connected systems for context.

3

Reason

An LLM processes the request using domain prompts, rules, and few-shot examples.

4

Act

Tool calls create records, route tasks, generate documents, or escalate to humans.

5

Log

Every decision, retrieval, and action is recorded in a complete audit trail.

Industry modules

Insurance

Claims, underwriting, policy servicing. Guidewire, Duck Creek.

Healthcare

Prior auth, clinical docs, patient intake. HL7 FHIR, Epic.

Finance

KYC/AML, risk analytics, loan processing. Core banking.

Manufacturing

Predictive maintenance, QC, supply chain. OPC UA, SAP.

Installation

Prepare your environment and deploy — from provisioning to live production in days.

Server requirements

ComponentMinimumRecommended
CPU4 vCPUs8+ vCPUs
Memory16 GB32 GB
Storage100 GB SSD500 GB NVMe
DatabasePostgreSQL 15+PostgreSQL 16 + pgvector
RuntimeDocker 24+Kubernetes 1.28+

Preparation

Provision a PostgreSQL instance with the pgvector extension, pull container images from the private registry, and configure SSO via OpenID Connect.

SQL
1# Provision database
2CREATE EXTENSION IF NOT EXISTS vector;
3CREATE DATABASE sphereiq;
shell
1# Pull container images
2docker pull registry.sphereiq.ai/platform:latest
3docker pull registry.sphereiq.ai/worker:latest
4docker pull registry.sphereiq.ai/ingestion:latest

Deploy with Docker Compose

For development, staging, or small-scale production:

shell
1export SPHEREIQ_DB_URL=postgresql://user:pass@host:5432/sphereiq
2export SPHEREIQ_LLM_PROVIDER=openai
3export SPHEREIQ_LLM_MODEL=gpt-4o
4
5docker compose up -d

Deploy with Helm (Kubernetes)

For production. Includes autoscaling, ingress, and monitoring:

shell
1helm repo add sphereiq https://charts.sphereiq.ai
2helm install sphereiq sphereiq/platform \
3 --namespace sphereiq \
4 --values values.yaml
Verify. After deployment, check curl https://your-instance.sphereiq.ai/health — all components should report healthy.

Configuration

All platform behavior is controlled through a single YAML file — tenants, data sources, prompts, and tools.

Configuration file

config.yaml
1# config.yaml — minimal example
2global:
3 llm_provider: openai
4 llm_model: gpt-4o
5 embedding_model: text-embedding-3-large
6 audit_enabled: true
7
8tenants:
9 - id: insurance-claims
10 industry: insurance
11 data_sources: [claims-kb, policy-docs]
12 human_review: true
13
14data_sources:
15 - id: claims-kb
16 type: hybrid
17 connection: postgresql://...
18 chunk_size: 512

Global settings

KeyTypeDescription
llm_providerstringopenai, azure, anthropic, or local
llm_modelstringModel identifier, e.g. gpt-4o
embedding_modelstringEmbedding model for vector search
audit_enabledboolFull decision audit trails (default: true)
max_tokensintMax response tokens (default: 4096)

Tenant isolation

Each tenant gets its own data sources, prompts, tools, and human-review threshold. Tenants share a deployment but cannot access each other's data. Setting human_review_threshold: 0.7 escalates any decision below 70% confidence.

Industry modules

Set industry on a tenant to auto-load domain prompts, tools, synonyms, and eval sets. Available modules:

  • insurance — claims, underwriting, policy, broker tools
  • healthcare — prior auth, clinical docs, FHIR connectors
  • manufacturing — maintenance, QC, OPC UA/MQTT ingestion
  • finance — KYC/AML, risk, regulatory reporting

Prompt templates

prompts/claims-intake.yaml
1# prompts/claims-intake.yaml
2role: |
3 You are a claims intake specialist.
4 Classify claims by severity and route to adjusters.
5
6rules:
7 - Never disclose internal routing logic.
8 - Cite the source document for any factual claim.
9 - Escalate when confidence is below threshold.
Hot reload. Prompt templates and synonyms maps are reloaded without restarting services. Global and tenant settings require a restart.

Security & Privacy

Built for regulated industries — SOC 2, HIPAA, PCI DSS controls in every layer.

All customer data is encrypted at rest (AES-256) and in transit (TLS 1.3). The platform undergoes annual penetration testing. Customer data is never used to train models.

Compliance

CertificationStatusScope
SOC 2 Type IIActiveAll services
HIPAABAA availableHealthcare
PCI DSSCompliantFinancial services
ISO 27001In progressAll services
GDPRCompliantEU data subjects

Data handling

  • No training on customer data. Your data is never used to improve any model.
  • Data residency. You choose the region. On-prem deployments keep everything in your infrastructure.
  • Retention. Configurable — default 90 days. Can be extended, shortened, or disabled.
  • Deletion. Complete deletion within 30 days on request, confirmed in writing.

Access control

  • SSO only. OpenID Connect via Entra ID, Okta, or any OIDC provider. No local passwords.
  • RBAC. Roles scope access to tenants, data sources, and admin functions.
  • Audit logging. All access events are logged immutably — logins, queries, config changes.

Supply chain security

  • Signed images. All containers are signed and verified before deployment.
  • Dependency scanning. Critical and high vulnerabilities block release.
  • SBOM. Software bill of materials available for every release.

Support

ChannelAvailabilityResponse
Email24/7< 4 hours
Slack ConnectBusiness hours< 1 hour
Phone (enterprise)24/7Immediate