Event-driven pub/sub
Endpoints publish and consume typed events through Azure Service Bus topics and subscriptions: one topic per endpoint, routed by event type.
Session-ordered delivery, failure recovery, the operator console, the CLI, and the extension framework, grouped by what you're trying to get done.
Publish typed events, route them by type, and process them in strict per-key order.
Endpoints publish and consume typed events through Azure Service Bus topics and subscriptions: one topic per endpoint, routed by event type.
Messages sharing a [SessionKey] are handled strictly
in order, one at a time. A stuck message blocks only its own
session.
Typed request/response conversations over Service Bus sessions, with timeout handling built in.
Schedule events for future delivery and cancel scheduled messages before they fire.
Declare endpoints in code, export the platform config, and provision topics, subscriptions, and routing rules from it, idempotently.
SQL Server or Cosmos DB behind one storage contract, with a conformance suite that keeps every provider honest.
An endpoint can opt in to publishing or consuming CloudEvents 1.0, in binary or structured JSON mode. Subscribers can detect the format, and sessions, retries, and tracking keep working.
Deriving a message from Command marks it as an
instruction with exactly one consumer. Topology provisioning
fails if a command has no consumer, or more than one.
Retries, dead-lettering, session blocking, and operator-driven recovery, with recorded outcomes for operators to investigate.
Retry policies per event type, exception type, or exception
text, with fixed, linear, or exponential backoff, optional
jitter, and support for server Retry-After hints.
Unrecoverable exceptions are classified and dead-lettered immediately, so no retries are spent on a payload that can never succeed.
When a message fails, later messages in the same session are deferred automatically until that session is unblocked at the receiving endpoint.
Once a session is unblocked, its deferred messages are re-published in FIFO order, so the backlog drains itself.
Failed events can be resubmitted or skipped from the WebApp while keeping the Resolver's history consistent.
Persist outgoing messages in a SQL Server outbox inside your business transaction; a dispatcher publishes them afterwards. The outbox ties the intent to publish to the database commit; consumers should remain idempotent.
An opt-in inbox records each successfully handled
MessageId per endpoint and skips a redelivered
duplicate instead of running the handler again. It stores to SQL
Server or Cosmos DB, with an in-memory store for tests.
An opt-in breaker stops an endpoint's receivers when handler failures pass a threshold, so queued messages stay on the subscription in order. It then probes with one session at a time and restores full concurrency once the probes succeed.
A handler can hand a message to an external job and return. The session stays blocked, in order, until the job reports back and the message is completed or failed, without the handler running again.
The operational layer that raw Service Bus doesn't give you: a queryable audit of every message, a console to act on it, and automation for the rest.
The Resolver records every message's outcome, such as Pending, Completed, Failed, Deferred, Dead-lettered, or Skipped, as one queryable audit trail.
An authenticated console for inspecting endpoints, events, and messages, with live updates, resubmit/skip actions, alerts, and audit views.
nb command-line toolProvision infrastructure, apply topology, deploy apps, purge queues, and clean up sessions, from a laptop or a pipeline.
Health probes for Service Bus and the message store, served on
/health, /alive, and
/ready.
Logging, OpenTelemetry metrics, and message validation ship as pipeline behaviors. Register the behaviors your application needs.
Failures, dead-letters, and session blocks fan out to notification channels: a webhook, Microsoft Teams, or email.
The WebApp sends a probe to every endpoint and the Resolver, and shows each one's status, round-trip time, and SDK version. Endpoints answer automatically, with no handler to write.
One page lists every unresolved failure across the endpoints you can read, with a failures-over-time chart, filters, and list, by-endpoint, and by-error views. Resubmit or skip from any view.
Reader, Contributor, and Owner roles apply site-wide and per
endpoint, managed on the Access Control page. A separate PII
Reader role sees raw values in [Sensitive] fields,
which are otherwise redacted, partly revealed, or hashed.
When a producer floods the bus, site Owners can find the backlog on the Topology page and stop or clear it per subscription without touching other endpoints. Every action is audited.
nb infra apply --network-mode private deploys
Service Bus Premium, the store, and both apps behind private
endpoints in your VNet, with public access off. It hasn't yet
been verified on a live Azure deployment.
You bring events and handlers; NimBus brings the hosting, pipeline, and a full local development story.
Register publishers and subscribers through DI, then configure
receiver hosting with AddNimBusReceiver for Workers
or Service Bus triggers for Azure Functions.
[SessionKey] sets the ordering scope; validation and
description attributes drive schema checks and documentation.
Wrap message handling with your own cross-cutting behaviors, using the same delegate-chain pattern as the built-ins.
Passive hooks for received, completed, failed, and dead-lettered events: observe without touching the pipeline.
Run the full pipeline, including ordering, retries, and outcomes, in unit tests with no Azure connection at all.
The CRM/ERP Aspire sample runs the provisioner, Resolver, WebApp, and adapters with the NimBus Service Bus emulator and SQL Server by default.
An adapter repository's Aspire AppHost can run the Service Bus emulator, topology provisioning, the Resolver, and the WebApp from NuGet tool packages, with no NimBus source checkout.
nb asyncapi exports the platform as an AsyncAPI 3.0
document, validates one, and diffs two, exiting non-zero on a
breaking change so it can gate a build.
nb catalog export writes a runnable EventCatalog
site.
A site Owner can turn on a Simulate page that publishes schema-valid generated events and stands in for chosen consumers, each with a failure mode. Traffic goes through the real bus and Resolver; production environments are always blocked.
A schema registry and REST agent API let AI agents discover event types and participate on the bus as first-class endpoints. An operator MCP endpoint lets agents monitor endpoints, investigate failures, and recover single failed messages under the WebApp's permissions and audit.
Extensions package up services, middleware, and observers behind one registration. Without any, NimBus runs with its default core behavior.
INimBusExtension model
Extension packages register services, pipeline behaviors, and
lifecycle observers through a single AddNimBus()
call, composed with a fluent builder.
Configurable notifications on lifecycle events (failure and
dead-letter by default) through webhook, Microsoft Teams, and
email channels, or your own INotificationChannel,
with per-channel severity filters and rate limiting.
ASP.NET Core Identity username/password sign-in for the WebApp, with email confirmation and reset flows, and optional dual login with Entra ID.
Off by default. An agent reads size-limited, redacted evidence of one failed or dead-lettered message over MCP and records its classification; NimBus calls no model. The result is stored with its history and comes with fixed guidance: retry may help, change likely required, investigate, or uncertain. It never changes message state.
AddAzureBearerToken adds a cached Microsoft Entra
bearer token to an IHttpClientFactory client, for
calls to Dataverse, Business Central, or Finance and Operations.
Run the CRM/ERP sample on your machine, or read how a message moves through the platform.