Inference Analytics
Also known as: inferenceanalytics.ai, Inference Analytics AI, InferAgents, Inferch, InferAnalyzer, PayerAgents, Inference Analytics Neural Network
A single platform, run under a business associate agreement, where a health system's own IT and informatics teams build, test and release AI agents on the models they choose. InferAgents is the visual studio, with retrieval over internal documents, PHI redaction, development, staging and production stages and a log of every prompt, retrieval and tool call. Inferch puts GPT, Claude and Gemini in one governed workspace, InferAnalyzer answers questions over SQL, Tableau, Excel and CSV data, and PayerAgents drafts cited denial appeals for billers to review. Pricing is quoted.
Inference Analytics sells one platform where a health system's own IT, informatics and engineering teams build, govern and run AI agents, in place of a separate point product for each problem. Four products share a core that operates under a business associate agreement. InferAgents is the agent studio.
Inferch puts models such as GPT, Claude Sonnet and Gemini side by side in one governed workspace, where a question goes to several models at once and a fusion answer shows which models contributed. InferAnalyzer answers plain language questions over SQL databases, Tableau, Excel and CSV files under the asker's own data permissions.
PayerAgents takes a CPT code, a denial reason and clinical context, finds the payer policy rule the denial turned on and drafts a cited appeal letter for a biller to check and send.
In InferAgents, builders assemble agents on a visual canvas from ready made stages: input, query rewriting, retrieval from vector stores and knowledge graphs, reranking, a model call, a hallucination guardrail and a cited answer, with enterprise APIs and tools available as steps. A healthcare retrieval flow uploads documents, redacts PHI, chunks the text and builds a vector store on the embedding model the team picks, and retrieval is filtered by the asking user's entitlements at query time.
Each step can run on a different model, from OpenAI, Anthropic and Google to open source, private and self hosted ones, and swapping a model does not mean rebuilding the agent. Context trimming, cheaper models for routine steps and usage budgets per user, team and agent hold down token spend. Agents start from an application, a chat or an API call, and they move through development, staging and production.
Governance runs through the shared core. An evaluation set is rerun before an agent or a new model is promoted, the person who builds an agent cannot approve its release, and an engineering owner fixes each agent's data and tool reach, which the agent cannot widen during a task. Review steps for outputs that inform care are set per agent category and cannot be removed in the builder, agents may read the EHR but never write to it, and submitting outside documents such as payer appeals stays with a person.
Every prompt, retrieval, tool call, model call and output lands in an immutable log tied to identity and time. Single sign on runs through the health system's identity provider, roles cover who can promote an agent to production, and data is encrypted with TLS 1.2 in transit and AES 256 at rest. The platform runs on major cloud providers that go unnamed, residency is agreed per deployment in the contract, and certification scope, the subprocessor list and the standard BAA are shared under NDA. Pricing is quoted after a demo, with no plans or rates published.
It fits health systems whose own engineering and informatics teams want to build many agents on one governed base, choose models step by step and keep PHI inside a single BAA boundary, and revenue cycle teams that want cited payer appeal drafts. It fits less well for buyers who want published pricing or a self serve trial, a named SOC 2 or HITRUST report before signing, an agent that writes back to the EHR or submits claims on its own, or memory that carries across sessions.
Vendor details
Canonical URL
https://www.inferenceanalytics.ai
Category
Healthcare agent
Subcategory
Platform where health system teams build, govern and run their own agents under one BAA, with a multi model workspace, an analytics agent and a payer appeals app
Funding status
Independent. Inference Analytics was founded in 2018, is based at the Merchandise Mart in Chicago and is led by founder and CEO Farrukh Khan. Funding is not stated on its own pages.
Company status
independent
Use cases & customers
Primary use cases
Target customers
Deployment options
Integrations
Agents reach enterprise systems through connectors and APIs confirmed for each deployment, with each tool's reach fixed when it is created. The Epic Tip Sheet Agent surfaces Epic tip sheets beside the system of record, and agents read the EHR under scoped access without writing to it. Knowledge comes from uploaded documents, vector databases, knowledge graphs and structured data. InferAnalyzer reads SQL databases such as SQL Server, Tableau, Excel, CSV and Parquet files, saved SQL connections and Fabric data agents. Models come from OpenAI, Anthropic and Google through their APIs under a BAA, alongside open source, private and self hosted models, with Azure Foundry shown as a model source.
In practice
An informatics team builds an agent that surfaces Epic tip sheets beside the record. It passes through development and staging, the evaluation set reruns, and a different person approves its release to production.
A biller enters an MRI denied as not medically necessary under CPT 72148. PayerAgents finds the payer policy criterion, drafts a cited appeal, and the biller checks and sends it through the usual channel.
Clinical staff compare answers from GPT, Claude and Gemini on one question in Inferch while IT controls which models each team may use, and an analyst asks InferAnalyzer about denial rates in Tableau without writing SQL.
Sources & related URLs
Related / legacy domains
Research sources
Agentic Index coverage score
10.0 / 14 capabilities · 71%
| Integrations & Tool Calling | Partial |
|---|---|
|
Agents reach enterprise systems through connectors and APIs confirmed for each deployment, and tools are added as steps on the canvas, each with its reach fixed when an engineering owner creates it. No connector catalog is named. Reading is the common case: agents read the EHR under scoped access and may not write to the patient record, and the Epic Tip Sheet Agent surfaces tip sheets beside the system of record. In PayerAgents the agent drafts the appeal and a biller submits it, so actions in outside systems stay with people. Sourceinferenceanalytics.ai/ai-governanceread 2026-11-14 |
|
| Workflow Orchestration | Full |
|
Agents are built as multistep flows on a visual canvas, chaining stages such as query rewriting, retrieval from vector stores and knowledge graphs, a reranker that decides which context to keep, a model call, a hallucination guardrail and a cited output. Steps can call tools and enterprise APIs, and each step can run on its own model, with routine steps routed to cheaper ones. Agents are versioned and promoted through development, staging and production, and one view lists every agent with its owner, data reach and model. Sourceinferagents.airead 2026-11-14 |
|
| Knowledge Grounding & RAG | Full |
|
The healthcare retrieval flow uploads documents, redacts PHI, splits text into 512 token chunks, generates hypothetical queries and builds a vector database on the embedding model the team chooses. Agents can also draw on knowledge graphs, documents and structured data, with a reranker and a guardrail stage before the answer, and answers carry citations. Retrieval is filtered by the asking user's entitlements at query time, and embeddings made from PHI are handled as PHI. PayerAgents answers each payer from its own policy library, with policies tracked daily. Sourceinferenceanalytics.ai/dev-studioread 2026-11-14 |
|
| Human Oversight & Guardrails | Full |
|
Review steps are set centrally for each agent category and cannot be removed in the builder, and a qualified reviewer is recorded for any output that informs care. Diagnosis, treatment selection, dosing, care plan changes and submitting outside documents such as payer appeals are reserved for people, and dismissing an agent's flag is logged against the person who did it. Building an agent and approving its release are separate roles, an agent cannot widen its own scope during a task, and PayerAgents drafts letters that a biller checks and sends. Sourceinferenceanalytics.ai/ai-governanceread 2026-11-14 |
|
| Security, Identity & Governance | Full |
|
Single sign on runs against the health system's own identity provider, with role mapping kept there, and roles cover platform functions down to who may promote an agent to production. Data and tool reach is set per agent, and retrieval follows each user's entitlements. Traffic uses TLS 1.2 or higher, including calls to model providers, data at rest uses AES 256 across stores, vector indexes, caches, backups and logs, and keys rotate and are never shared across tenants. A BAA is signed before any PHI enters and extends to model providers. No SOC 2, HITRUST or ISO certificate is named publicly, and certification scope and assessment dates are shared under NDA. Sourceinferenceanalytics.ai/securityread 2026-11-14 |
|
| Observability & Auditability | Full |
|
Every prompt, retrieval, tool call, model call and output is logged with the user's identity and a timestamp. Logs are immutable, can be queried and are exportable for a named user and date range, and reviewer identity is recorded alongside. Usage and token spend show per user, team and agent, and one view lists every agent with its owner, data reach and model. Model changes carry an owner, a reason and a timestamp, and rollback keeps the prior configuration available. Sourceinferenceanalytics.ai/securityread 2026-11-14 |
|
| Memory & State Persistence | Not documented |
|
Nothing carries facts about a user, a patient or a workflow from one session into the next. Agents ground their answers in document stores and knowledge graphs that the team maintains, which are the organization's knowledge and not a record of past conversations. In PayerAgents, conversations are deleted after each session and practice data is never used for training, while Inferch and InferAnalyzer answer each question against current sources. Sourceinferenceanalytics.ai/payeragentsread 2026-11-14 |
|
| Deployment & Data Residency | Partial |
|
The platform is set up per organization on major cloud providers that are not named, and each customer's data is logically isolated with no retrieval across tenants. Residency is agreed for each deployment and written into the contract, and specific regions are shared during evaluation. Workloads marked internal only can run on the health system's own internal or self hosted models. VPC, on premises and customer tenant deployment of the platform itself are not described. Sourceinferenceanalytics.ai/securityread 2026-11-14 |
|
| Prebuilt Agents / Templates / Packs | Full |
|
Several ready made agents ship alongside the studio. PayerAgents bundles a payer policy library chat, a CPT checker, an appeal letter builder and an eligibility checker for billing teams. Inside Inferch, InferAnalyzer answers questions over SQL, Tableau, Excel and CSV data, a presentation agent drafts slide decks from a prompt and a template, and advisor agents reason as expert personas with domain context. The studio itself starts builders from ready made stages, such as the healthcare retrieval flow. Sourceinferenceanalytics.ai/inferchread 2026-11-14 |
|
| Triggers & Channel Coverage | Partial |
|
An agent starts when an application calls it, when someone chats with it, or through an API call. PayerAgents runs when a biller picks a payer and enters a denial, Inferch and InferAnalyzer answer questions as staff ask them, and the Epic Tip Sheet Agent responds beside the record. No schedule, webhook, shared inbox or EHR event starts an agent on its own in any of the four products, so every run begins with a person or an application asking. Sourceinferagents.airead 2026-11-14 |
|
| Model Flexibility & Routing | Full |
|
Teams choose the model for each task by quality, latency and cost, from OpenAI, Anthropic and Google models to open source, private and self hosted ones, and switch models without rebuilding the agent. Temperature, max tokens and top P are set per model step. Administrators decide which models are approved, for whom and within what limits, and new models reach selected agent categories first under change control. In Inferch, staff send one question to several models, such as Claude Sonnet, GPT and Gemini, and read a fusion answer that names the models behind it. Sourceinferenceanalytics.ai/platformread 2026-11-14 |
|
| APIs / SDKs / MCP Extensibility | Partial |
|
Agents can be started by an API call, and the studio connects to enterprise APIs as steps inside an agent. No public API reference, SDK or MCP server for the platform itself is offered, and the setup of connections comes through the deployment team. Connecting a new source system and approving a production release remain engineering work for the customer's own team, while common agents are built without code. Sourceinferagents.airead 2026-11-14 |
|
| Testing, Debugging & Optimization | Full |
|
Agents carry an evaluation set that is rerun before any change is promoted to production. Candidate models are tested against the same evaluation set used by the agents that might adopt them, then introduced to selected agent categories first, with rollback to the prior configuration. Agents are tested and versioned through development, staging and production, and a hallucination guardrail stage sits in the flow before an answer goes out. Ongoing monitoring of agents in production is not described in detail. Sourceinferenceanalytics.ai/ai-governanceread 2026-11-14 |
|
| Browser / Computer-use | Not documented |
|
Agents work through retrieval, connectors, APIs and the data sources InferAnalyzer reads. None of the four products drives a web browser or a desktop application, and nothing describes agents clicking through payer portals, EHR screens or other interfaces on their own. PayerAgents prepares appeal letters that a biller sends through the usual channel, and the Epic Tip Sheet Agent shows guidance beside the record without operating the EHR. Sourceinferenceanalytics.ai/payeragentsread 2026-11-14 |
|
The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Pricing
Contact sales
What is public
No prices, plans or billing basis are published for InferAgents, Inferch, InferAnalyzer or PayerAgents. A demo request is the route in, each deployment is set up per organization, and a BAA is signed before any PHI is involved.
Variable cost rationale
No billing unit, rate, minimum or cap is published, so whether the bill grows with users, agents, queries or model tokens is not stated. Model calls run on OpenAI, Anthropic, Google and self hosted models, and how those token costs reach the customer is not described.
Sales call required
Yes, required for paid access
Free / trial
None published
Lowest paid plan
Not published
Key ambiguities
Whether model token costs are billed through Inference Analytics or on the health system's own provider accounts is not stated. The cost claim appears as 'up to 80% lower Gen AI costs on qualifying workloads' on the Inferch page and '80%+ token savings on benchmarked workloads' on inferagents.ai, neither with a baseline.
Related vendors
- Abridge — AI platform for health systems that turns clinical conversations…
- Adonis — AI orchestration platform for revenue cycle management whose…
- AKASA — Generative AI for the hospital mid revenue cycle: Coding and CDI…
- Alaffia Health — Agentic AI for health plan claims operations: utilization…
- AlethianAI — Connected AI agents for medical practices share one patient record
- Ambience Healthcare — AI platform for health systems pairing an ambient scribe with…