Our experience has shown that building high-quality conversational agents requires extremely close collaboration between developers and business teams. When proof-of-concepts fail, it’s rarely a technology problem — it’s usually an iteration issue.
Today, even simple changes — like adjusting temperature, tweaking default agent personality via the system prompt, or providing additional knowledge into the context window — require code changes and redeployments. This turns small iterations into days- or weeks-long feedback cycles.
At the same time, existing observability and evaluation tools operate at a very low technical level. They make it hard for non-technical users to follow real conversation flows, debug issues, or validate whether an agent behaves as intended.
We started Hebo to bridge this gap — and to enable fast, shared iteration between product, business, and engineering teams. Hebo is:
- A unified AI gateway providing instant access to state-of-the-art LLMs with low-latency, high-throughput completions and embeddings.
- A composable stack with primitives for RAG, Tools, and Memory that give fine-grained control over conversation content and quality.
- An evaluation & observability toolchain designed to make real conversation flows visible, debuggable, and testable — for both technical and non-technical teams.
All this wrapped into a simple user interface targeted at operational teams to define agent behavior, observe conversations and optimize knowledge.
Every part of Hebo is designed to be opt-in, with a natural ramp-up as your agent evolves in sophistication — from simple prompts to fully-fledged, production-grade helpers.
Hebo is built in the open with an FSL license that keeps Hebo free to use for everyone while discouraging harmful free-riding. We’re still at the very beginning. Follow us on X (@heboai) to stay up to date, or reach out to us on Discord to join the conversation.
Get started on the Hebo Console, browse the docs, or check out the project on GitHub.