Show HN: Inkeep (YC W23) – Agent Builder to create agents in code or visually (github.com)
As a developer, the flow is: 1) Build AI Chat Assistants or AI Workflows with the TypeScript SDK 2) Run `inkeep push` from your CLI to publish 3)Edit agents in the visual builder (or hand off to non-technical teams) 4) Run `inkeep pull to edit in code again.
We built this because we wanted the accessibility of no-code workflow builders (n8n, Zapier), but the flexibility and devex of code-based agent frameworks (LangGraph, Mastra). We also wanted first-class support for chat assistants with interactive UIs, not just workflows. OpenAI got close, but you can only do a one-time export from visual builder to code and there’s vendor lock-in.
How I've used it: I bootstrapped a few agents for our marketing and sales teams, then was able to hand off so they can maintain and create their own agents. This has enabled us to adopt agents across technical and non-technical roles in our company on a single platform.
To try it, here’s the quickstart: https://go.inkeep.com/quickstart.
We leaned on open protocols to make it easy to use agents anywhere: An MCP endpoint, so agents can be used from Cursor/Claude/ChatGPT A Chat UI library with interactive elements you can customize in React An API endpoint compatible with the Vercel AI SDK `useChat` hook Support for Agent2Agent (A2A) so they work with other agent ecosystems
We made some practical templates like a customer_support, deep_research, and docs_assistant. Deployment is easy with Vercel/Docker with a fair-code license and there's a traces UI and OTEL logs for observability.
Under the hood, we went all-in on a multi-agent architecture. Agents are made up of LLMs, MCPs, and agent-to-agent relationships. We’ve found this approach to be easier to maintain and more flexible than traditional “if/else” approaches for complex workflows.
The interoperability works because the SDK and visual builder share a common underlying representation, and the Inkeep CLI bridges it with a mix of LLMs and TypeScript syntactic sugar. Details in our docs: https://docs.inkeep.com.
We’re open to ideas and contributions! And would love to hear about your experience building agents - what works, hasn’t worked, what’s promising?
18 comments
[ 2.5 ms ] story [ 35.2 ms ] thread> We built an agent builder with true two-way sync between code and a drag-and-drop visual editor.
Wow, what a clear pitch. I like it.
At the same time, I think about design space between Visual/DAG editors (here, a directed graph of agent workflows) versus, say, a high level textual configuration format (a la Dockerfiles).
- I think back ... how many visual tools have I been excited by [2] [3] [4] [5] [6], only to find that I usually prefer the textual editing most of the time? There are certainly cases where the visual editors really catch on. But on the other hand, when it comes to the programming world, it seems like the configuration format approach works more often.
- What do customers want here? (I don't have any particular expertise here) In my footnoted examples, my guess is that visual tools catch on the best when the target audience has a deep physical, even tactile, connection to the domain rather than a preference for textual representations.
Personally, I really like both. I like being able to quickly edit and share text files and also switch to a visualization. But it can be hard to make the visualization capture the necessary details without too much clutter.
All in all, delivering on two-way sync between code and visual editors might be hard. Hard is not necessarily bad. Delighting customers on both fronts could be a competitive advantage, for sure. [7]
--
I know this comment could be better organized, sorry about that. This is a "thinking out loud comment"... I haven't even touched on the "no code" and "low code" angle to it. I'd be happy to hear from others on their experiences.
[1]: https://www.youtube.com/watch?v=4FuEnAEPqwU
[2] Tools like SAS Enterprise Miner (https://www.sas.com/en_us/software/enterprise-miner.html) or Orange Data Mining: Visual Programming: (https://orangedatamining.com/home/visual-programming/)
[3]: Max for Live (integrated with Ableton for sound design)
[4]: LabVIEW (used for electrical engineering)
[5]: Various visual SQL Schema editors
[6]: Graphical views of document linkages: e.g. Obsidian, The Brain (going way back)
[7]: It may be difficult in achieve parity between the different capabilities of each. It seems to me many applications recognize that full parity isn't practical and instead let each "view" do what it does best. Traditionally, the visual approaches help with the top-level view and the code versions get into the details.
inkeep isn't open source with a Elastic License 2.0, why not just go with OpenAI agents sdk (MIT)?
The visual UI + code is really cool, addresses the weaknesses of both approaches.
-> connect to the locally installed instance of SigNoz
-> It asks for an email -> when I type it, it says: "This account does not exist. To create a new account, contact your admin to get an invite link"
-> But I am the admin :'), also tried to create an account there https://signoz.io/
-> but they refuse personal Github or Gmail accounts for now.
Conclusion
So it's literally impossible to run your app for a 'normal person' running their own server :(
Or maybe I missed a step :/
(ok you guys, we've taken 'open source' out of the title, let's talk about something more interesting now)
This might sound vaguely pessimistic, but I'm getting the nagging feeling lately that we might be inflating a ponzi scheme of B2B saas selling exclusively to other B2B saas companies for niche B2B saas use cases.
In earlier generations of Saas the assumption was you land and expand from there to wider markets. But this seems so specifically tailored to the needs of other VC-funded B2B Saas companies...
As the things that fall under [vibe code-able in 1 hour] expands, at what point does building platforms like this for semi-technical B2B Saas employees not make sense any more?
Like, if you're smart enough to figure out an agent that connects the OpenAI API to the Zendesk API can automate something...at what point do you just set it up yourself instead of dealing with the sales vultures at [insert YC startup] and also fighting your own internal procurement process...just so you might be able to have that agent running in 1-6 months?
In enterprise I can totally see value in having help during setup and an external throat to choke when things go wrong, but at that point, isn't that a consulting agency masquerading as a Saas platform?
EXACTLY what was missing in many tools, and exactly what is required for reducing the time from quick prototype to production quality code.
Congrats on the launch. Looking forward to trying this out. Please add support for AWS Bedrock based models.
automatically rewrite the TS file on the next pull, or surface a merge conflict and let the dev hand-fix? Curious how deep the refactor detection goes.
I recently reviewed n8n and Dify for this (https://www.cyprien.io/posts/n8n-vs-dify/), but I will definitely check this one.
How do you compare yourself to Dify?