AgentKit and Inngest are awesome. Thank you @tonyhb and the Inngest team for building such an amazing DX.
I've been testing Inngest realtime with AgentKit and it's an awesome combo already despite being a few weeks old.
I highly recommend the Inngest demo checking out their SWE bench example and the E2B example (https://github.com/inngest/agent-kit/tree/main/examples/e2b-...) to build a network of coding agents. I've also got this working with Daytona sandboxes.
Some feedback: - Would love an llms.txt for your docs like https://developers.cloudflare.com/agents/llms-full.txt - clear way of accessing realtime publishing from within AgentKit.
You can grab the LLMs-full.txt file here: https://agentkit.inngest.com/llms-full.txt
More realtime examples coming soon!
I've been using Agentkit + Inngest to build out an agentic customer support network for the past week and the experience so far has been wonderful.
I've been evaluating n8n and Mastra.ai at the same time to determine the best platform for my use-cases and Agentkit + Inngest has been the clear winner IMO.
The fact that Agentkit is able to leverage Inngest's durable workflow execution engine is awesome as it makes the interaction with the agentic network waaaayy more reliable.
How's the performance compared to LangGraph? I'm working on a project that needs to handle high throughput agent interactions.
Akka focuses on enterprise agentic with a focus on creating certainty and solving scale problems. We have a customer, Swiggy, which is >3M inferences per second for a blended set of models, both ML and LLMs, with a p99 latency of roughly 70ms.
This level of throughput is achieved by including memory database within the agentic process and then the clustering system automatically shards and balances memory data across nodes with end user routing built in. Combined with non-blocking ML invocations with back pressure you get the balance for performance.
The framework itself is super low overhead. You can deploy this anywhere, and if you deploy to inngest.com the P99 latency of starting agents is sub-50ms (and you can also realtime stream steps, tools, or model responses to the browser).
One of the main differences is the DX — _how_ you define the agentic worklflows is far cleaner, so it's both faster to build and fast in production.
Great stuff. Would be neat to have an adapter make the output stream match the ai-sdk format so it plugs in to existing UX (https://sdk.vercel.ai/docs/ai-sdk-ui/stream-protocol)
Interesting, have you considered adding benchmarks comparing AgentKit to other frameworks? Would help teams evaluating options
This looks really good. I'm going to take a detailed look at this for sure. Thanks!
Really love the decoupling of the logic and the runtime for the actual tool calls.
That is great! How does debugging work with the local DevServer?
If you use AgentKit + Inngest then you can do all the things you normally do with the Inngest dev server like observe runs with AI metadata and rerun functions and rerun functions from steps with edited inputs. We do have a step-through debugger coming pretty soon here as well. Note you don't have to use AgentKit with Inngest though.
good stuff
[flagged]
The main one is deterministic routing (https://agentkit.inngest.com/advanced-patterns/routing). Here's what that means:
Each agent builds up state via tool use. On each loop of the network, you inspect this state to figure out which agent to run next. You don't build DAGs or create odd graphs — you write regular code in a router.
Or, more generally:
* Each agent has a specific goal within a larger network. Several agents each working on smaller goals means easier prompt generation, testing, iteration, and a higher success rate.
* The network combines agents to achieve an overall objective, with shared state modified by each agent
* The network’s router inspects state and determines which agent should run next
* The network runs in a loop, calling the router on each iteration until all goals are met
* Agents run with updated conversation history and state on each loop iteration
Realistically the challenge with agents has classically been: how can I build something reliable, and how can this run in production reliably? These patterns are largely what we've seen work.
Thanks for explaining. But I guess I'm still not clear on how the work gets divvied up. Not tryin to be a hater - I'll have to give it a spin - but that part's a bit murky to me still.