Leah AI

A browsable connector catalog across eight categories, ERP finance and P2P, document management and e-sign, collaboration, CRM, HR payroll and talent, IT and DevOps, analytics, and translations, naming SAP, Oracle, Workday, NetSuite, Salesforce, Dynamics, Coupa, Ariba, Jaggaer, iValua, ServiceNow, DocuSign, Adobe Sign, ADP, BambooHR, Amazon S3, BigQuery and more, alongside open APIs for custom integrations across the stack. Agents take authenticated action in those systems rather than reading from them: the documented workflow generates a contract, routes it to DocuSign, notifies finance and starts obligation tracking, and agents run signature requests and approval routing.

SourceLeah, leahai.com/products/integrations and leahai.com/products/agentic-osread 2026-09-07

Skill based orchestration is documented as the platform's architecture: work routes to the right agents, they run in parallel across legal, contracting, procurement and finance with shared context, a procurement event cascades to contract and legal agents and on to finance obligations, exceptions escalate automatically, and the execution engine plans, acts, coordinates and self recovers when a process changes mid run. Agentic Workflows add a visual drag and drop builder encoding dependencies, approvals, escalation paths and regulatory constraints while maintaining state over time.

SourceLeah, leahai.com/products/agentic-osread 2026-09-07

The Enterprise Context Layer is documented as the substrate every agent reasons against, a maintained structure over the customer's own knowledge holding structured data (contracts, CLM records, POs, invoices, suppliers, approvals), enterprise knowledge (policies, playbooks, precedents, standard clauses, history) and live system data from ERP, CLM, procurement and identity feeds, connected to the live business rather than a static snapshot. It persists, scales past a run and stays queryable by every agent.

SourceLeah, leahai.com/products/agentic-osread 2026-09-07

The vendor's own review and approve surface is documented at three levels: governance rules the customer sets route work by threshold, shown as medium and above risk routing to procurement review; exceptions escalate to the named owner automatically with context pre assembled, with approval routing awaiting GC review in the live agent view; and Live Work Portals provide dashboards, intake and review for the humans in the loop. The FAQ states structured policy governed work runs autonomously while judgment critical decisions escalate with full context, and humans set policy and approve escalations.

SourceLeah, leahai.com/products/agentic-osread 2026-09-07

Attestations and access controls are both documented on the platform page. The compliance list covers SOC 1 Type 1 and 2, SOC 2 Type 1 and 2, ISO 27001:2022, GDPR, CCPA and HIPAA readiness, and the vendor states it supplied comprehensive security documentation to a financial institution's security teams. Role based access is named in the governance layer alongside compliance guardrails and secure access controls, and data is protected with Azure Key Vault key management, TLS in transit, AES-256 at rest and zero data retention with the model providers.

SourceLeah, leahai.com/products/agentic-osread 2026-09-07

The platform is described as audit native: every action traceable from trigger to outcome, every deviation documented, every approval logged with context, tamper resistant audit logs, clear decision logs, and a governed output the vendor calls board ready and regulator ready. The workflow memory layer carries agent history, cross agent context and the audit log, and the live orchestration view shows each agent's step and state. This is a runtime record of what the agents did and why, not reporting and not logs supplied by the customer's own systems.

SourceLeah, leahai.com/products/agentic-osread 2026-09-07

The architecture names Workflow Memory as a layer holding in flight state, agent history and cross agent context, and lists agent memory as a component of the execution engine. Scope is stated, per workflow and shared across agents in a run. No lifetime is stated and nothing documents where the memory lives or how a buyer deletes it for one customer. The compounding claim that every process run embeds more context describes absorbed learning rather than a memory layer, and the same layer also holds the audit log.

SourceLeah, leahai.com/products/agentic-osread 2026-09-07

The platform page states the customer chooses the deployment model that meets its security requirements, Leah's cloud, a private cloud, or fully on premise, with configurable safeguards by region, process and vendor type, and repeats flexible deployment across cloud, private cloud and on premise under enterprise trust; the FAQ names private cloud and on premise options again.

SourceLeah, leahai.com/products/agentic-osread 2026-09-07

Leah ships pre built domain agents across legal, contracting, procurement and finance, already trained on how the work flows and customizable to the buyer's policies and approval chains: contract review agents applying a playbook, supplier qualification agents triggering legal review on risk flags, obligation tracking agents alerting before deadlines, and regulatory response agents finding every affected contract.

Pre built agentic components are listed as assemblable units, and the platform is sold in named products, Leah Legal, Leah Procurement, Leah Finance and Leah Agentic CLM. Each agent does its own job when selected.

SourceLeah, leahai.com/products/agentic-osread 2026-09-07

Work reaches the agents on several documented event classes rather than by invocation: a procurement event automatically updates contract and legal agents, a contract change cascades to finance obligations in real time, a detected regulatory update fires across the affected contract portfolio and supplier records, exceptions trigger escalation to the right person, and obligation tracking agents alert before deadlines, which is time based. Live Work Portals add an intake surface, and connected systems feed events from ERP, CLM and procurement.

SourceLeah, leahai.com/products/agentic-osread 2026-09-07

The platform page states Leah is powered by multiple large language models and dynamically selects the ideal model for each task, whether a clause review, a regulatory risk assessment or a supplier qualification check, and names OpenAI, Anthropic and Azure OpenAI among its integrations with zero data retention at those providers. The routing is the vendor's, and no customer or admin control of the model is documented; the no code building described alongside it composes workflows, not model selection.

SourceLeah, leahai.com/products/agentic-osread 2026-09-07

Open APIs for custom integrations across the customer's full stack are named on the integrations page and in the platform's connector list, but no endpoints, resources, authentication or developer portal are published, and no access path to a reference is documented beyond a demo request.

SourceLeah, leahai.com/products/integrations and leahai.com/products/agentic-osread 2026-09-07

The execution engine lists Evaluation and Optimization as a layer, naming continuous eval loops alongside agent memory and audit native logging, and the platform is described as self healing, correcting mid execution when a supplier fails qualification or a regulation shifts. Nothing about the loop is documented beyond the label, with no harness, scored cases, release gate or readable outcome, and self correction at runtime is recovery rather than evaluation of a change.

SourceLeah, leahai.com/products/agentic-osread 2026-09-07

Leah sits above the customer's systems and reaches them through pre built connectors and open APIs, a positive statement that execution runs through integrations rather than an interface an agent operates. No browser, desktop or remote computer control is documented; the embedded interfaces named are the vendor's own portals and workflows.

SourceLeah, leahai.com/products/agentic-os and leahai.com/products/integrationsread 2026-09-07