Foundry hands you Azure-locked building blocks — a model service, a database, a search index, a gateway — to assemble and operate into an agent. Akka delivers agents, memory, streaming, APIs, and governance as one runtime you can run on any cloud, on-prem, or in a sovereign region. Foundry leaves the integration and the recovery across your Azure estate for you to own. Akka is built, operated and portable.
A production agent on Foundry is stitched together from separately provisioned Azure services. Microsoft's Standard setup requires you to bring and operate a Cosmos DB account, an AI Search resource, an Azure Storage account, and a Key Vault; a realistic agent also draws in Azure OpenAI, an API Management gateway, and Azure Monitor. You integrate them, and you own the reliability and recovery across every seam.
The assembled services run on Azure identity, gateway, and data. Foundry provides no automatic failover or disaster recovery for the Agent Service that ties them together, and moving the system off Azure means rebuilding it.
Akka gives you agents, durable memory, real-time streaming, an HTTP and gRPC API layer, orchestration, observability, and governance as one pre-integrated runtime — provisioned once and operated as one system. You describe the system you want, and the platform generates it, tests it, and runs it.
Akka runs on any cloud, on-prem, or in a sovereign region, with portable specs and no lock-in to one provider's identity or data services. Akka keeps running with active-active replication and automatic failover under a contractual 99.9999% availability SLA on the entire platform.
Akka Optimize puts each request on the best model for the task, including smaller models trained on your data.
| Dimension | With Foundry | With Akka |
|---|---|---|
| Scope | An agent service that orchestrates a set of separately provisioned Azure services — Cosmos DB, AI Search, Storage, Key Vault, OpenAI, API Management, Monitor — which you integrate and operate yourself. | Agents, memory, streaming, APIs, and governance on one pre-integrated runtime, provisioned once and operated as one system. |
| Availability | Foundry provides no automatic failover or built-in disaster recovery, and the Agent Service has no supported active-active multi-region replication, per Microsoft's documentation. | 99.9999% on the entire platform — about 31 seconds a year — active-active, sub-1-minute RTO, zero-byte RPO, backed by contractual indemnities. |
| Recovery point | Recovery reconstructs a warm standby. The documented recovery point for agent state "can be total loss," and standby-region state is lost on failback. | Zero-byte RPO — durable state is active-active replicated and fully preserved across a regional event. |
| State & memory | Agent Service long-term memory is a managed feature in preview. Agent state and conversation threads live in a Cosmos DB account you bring, size at a 3,000 RU/s minimum, and operate — including its backup, replication, and failover. | Durable state held in memory, active-active replicated — 4 ms reads, sub-10 ms writes — built into the runtime. |
| Governance | Guardrails check agent inputs, tool calls, tool responses, and outputs (agent support in preview) for agents built in Foundry Agent Service only. Purview Compliance Manager has assessment templates for the EU AI Act, ISO/IEC 42001, ISO/IEC 23894, and NIST AI RMF. | Akka Verify applies any control to any AI system and guarantees its enforcement — HITL, HOTL, guardrails, evaluations — into tamper-evident storage. |
| Spec-driven delivery | GitHub's open-source Spec Kit takes a project from a written specification through a plan and tasks to implementation. Foundry scores agents with rubric evaluators (preview). | Akka Specify builds agentic systems from your specifications and releases only what meets your definition of done. |
| Model optimization | Model router sends each prompt to one of 35 models from OpenAI, Anthropic, xAI, DeepSeek, Meta, and Fireworks. Fine-tuning supports supervised, preference, and reinforcement training. | Akka Optimize puts each request on the best model for the task, including smaller models trained on your data. |
| Portability | Built on Azure identity, gateway, and data services; the system is bound to Azure, and moving off means rebuilding it. | Any cloud, on-prem, or sovereign region — portable specs, one runtime, no lock-in to a single provider's identity or data services. |
| Infrastructure cost | The agent service plus separately provisioned Cosmos DB, AI Search, Storage, Key Vault, OpenAI, API Management, and Monitor — each sized with its own headroom and on its own meter. | One shared-compute runtime carries the same volume on up to 90% less infrastructure, at a fixed annual fee. Akka Optimize keeps improving the models on your own production traffic, so cost per task keeps falling as the system runs. |
| Continuous intelligence | Model routing, evaluation, and fine-tuning are separate Foundry capabilities, each on its own meter. | Akka Optimize runs one loop in the runtime that grades your production traffic, trains smaller models on your own data, and keeps cost per task falling as it runs. |
To get the same capabilities of Akka, you need to integrate numerous, high-margin, meter-based services. Akka delivers higher performance and lower costs by only using Azure commodity infrastructure.
| Azure chargeable item | Without Akka | With Akka |
|---|---|---|
| Foundry Agent Service | Per vCPU-hour and GiB-hour (hosted agents), per tool session or transaction | Included |
| Foundry Models | Per token | |
| Azure AI Content Safety | Per text record | |
| Azure AI Search | Per search unit-hour | |
| Azure Cosmos DB | Per request unit, per GB | |
| Azure Cache for Redis | Per node-hour | |
| Logic Apps, Durable Functions | Per action executed | |
| Service Bus, Event Hubs, Event Grid | Per throughput unit, per message | |
| Azure Stream Analytics | Per streaming unit-hour | |
| Azure API Management | Per unit-hour | |
| Azure Monitor, Application Insights | Per GB ingested | |
| Foundry evaluations | Per evaluation | |
| Azure Machine Learning, Foundry Models fine-tuning | Per training token or hour, per hosting hour | |
| AKS | N/A | Per cluster-hour |
| Azure Virtual Machines | N/A | Per instance-hour |
| GPU virtual machines | N/A | Per accelerator-hour |
| Azure Database for PostgreSQL | N/A | Per instance-hour plus storage |
| Azure Blob Storage | N/A | Per GB stored and per request |
Akka provides spec-driven delivery, guardrails, red teaming, evaluations, AI policies, AI orchestration, AI memory, AI routing, inference, and SLM training within a single integrated stack that depends upon commodity infrastructure provided by the hyperscaler. Akka charges a fixed, annual fee without metering. With the hyperscaler, you integrate various high margin services together to create an AI solution. Each of those metered services include the relevant compute, storage, and I/O infrastructure within their fees.
Foundry guardrails block flagged content in the inputs, tool calls, tool responses, and outputs of agents built in Foundry Agent Service.
Block, stop & assess
Foundry guardrails check an agent's input, tool calls, tool responses, and output, and block content that breaks a control. Guardrails for agents are in preview and apply only to agents built in Foundry Agent Service. Control Plane lets an administrator stop a hosted agent or block requests to a registered agent.
Purview Compliance Manager has assessment templates for the EU AI Act, ISO/IEC 42001, ISO/IEC 23894, and NIST AI RMF, and syncs 15 automated Foundry evaluation results into them. Purview captures agent prompts and responses in its unified audit log.
Block & escalate
Risk teams govern every agentic system without waiting on its developers, and every control they apply is guaranteed to be enforced. Risk teams choose from 1,718 controls drawn from 241 regulations and frameworks. Human-in-the-loop approvals, human-on-the-loop halt switches, sanitizers, guardrails, evaluations, testing gates, and simulations all run before an action executes and block or escalate anything that does not pass.
Every call is traced automatically into petabyte-scale, tamper-evident storage, and that same trace data feeds continuous evaluations that keep running in production, right alongside your agents.
Reach for Foundry when your organization is committed to Azure and wants to build against the Azure OpenAI models and the surrounding Azure services, with a team ready to provision, integrate, and operate Cosmos DB, AI Search, Storage, the gateway, and monitoring — and to own the availability and recovery of the assembled agent yourself.
Reach for Akka when you want agents, memory, streaming, APIs, and governance pre-integrated and operated under one SLA, when a regional outage cannot mean losing agent state, and when you need to run on any cloud, on-prem, or in a sovereign region without binding the system to a single provider's identity and data services.
Foundry gives you Azure-locked building blocks to assemble and operate. The Akka Agentic AI Platform builds, runs, governs and optimizes the whole agentic system under a contractual SLA, on any cloud, on-prem, or in a sovereign region. Tell us what you are building, and we will show you how it runs in production.
