turisanapo

Building

[Pi @pidotdev]

Pi now has 100000 stars on GitHub⭐️Thank you People of Pi for your contributions and support. The community is growing faster than ever!

Pi v2 coming soon…

[turisanapo @turisanapo]

Hey @saltyAom, just bumping this in case it got lost in the noise

[turisanapo @turisanapo]

Hey @saltyAom, some work to make @elysiaJS opentelemetry plugin safer by default:

https://t.co/rNgu5y8beB

[turisanapo @turisanapo]

Docs PR is here too:
https://t.co/oM210Kdy6u

[Heinrich Wendel @hmw147]

let’s make Elysia the main framework for MCP servers

https://t.co/bWyuqriPlB

@saltyAom

[Heinrich Wendel @hmw147]

MCP sounds complex, but it’s essentially just like OpenAPI. If you have existing API routes you should be able to expose them to agents with one line.

https://t.co/yENVH70YSI

@elysiaJS

[Heinrich Wendel @hmw147]

MCP shouldn’t mean rewriting your API.

elysia-mcp turns existing ElysiaJS routes into MCP tools.

Just add `.use(mcp())`:
→ Based on your existing Schema
→ Reuse OpenAPI-style metadata
→ Keep your handlers untouched

🐒 https://t.co/TvLea27kgm

@saltyAom @elysiaJS

[turisanapo @turisanapo]

@heiwen @saltyAom @elysiaJS When an API already exists, the easiest way for AI agents to work with it is through MCP.

This plugin provides the missing bridge between Elysia routes and MCP tools.

[turisanapo @turisanapo]

@heiwen A big part of the direction here is building a reliable interface for stateful AI APIs over time. Conversations were the first piece, /responses is the next one, and the real value comes when they work together cleanly.

[Heinrich Wendel @hmw147]

Hebo Gateway v0.9 → OpenResponses API support.

Embed a /responses endpoint directly in your own API.
Own the infra. Own the data. Full control. 🐒

https://t.co/xYDdeILEjw

[turisanapo @turisanapo]

@heiwen @Greptime Probably the part I’m most excited about. As soon as /responses becomes stateful, managing conversation state for agents gets much more straightforward.

[Heinrich Wendel @hmw147]

Most LLM infra treats calls as stateless.

But real apps need history, memory, traceability 🍌

This is our first step toward fully stateful /responses:
💬 /conversations API (OpenAI-compatible)
🪣 Pluggable storage (memory, Postgres, SQLite, MySQL, @Greptime )
🧠 Support for latest GPT-5.3 / 5.4 models
🐛 Fixed Amazon Nova model integration

https://t.co/NOd9HVmnFj

[turisanapo @turisanapo]

@heiwen @Greptime @ClickHouseDB Greptime has a solid SQL interface, strong ingestion, and a modern architecture that feels more approachable than many observability databases. The harder part was production setup, especially persistence, WAL, and cluster design for a small team. Still worth it.

[Heinrich Wendel @hmw147]

Most observability stacks look like: SDK → queue → workers → transformation → database

Ours: SDK → GreptimeDB

Why we chose @Greptime over @ClickHouseDB for AI observability 👇

https://t.co/zY6QswveKR

[turisanapo @turisanapo]

@heiwen This is especially nice for non-time-sensitive OCR jobs. You can just switch to flex and save ~50% without dealing with provider-specific quirks.

[Heinrich Wendel @hmw147]

🐒 Hebo Gateway 0.7

We just added standardized Service Tiers across: OpenAI, Bedrock, Groq, Vertex

The hardest one? Gemini. It exposes tiers via HTTP headers instead of request fields.

Hebo normalizes all of this so you can just write: `serviceTier: "priority" / "flex" / "scale"`

…and it works across providers 🍌

https://t.co/V1QlY29qhZ

[turisanapo @turisanapo]

@heiwen Nice. 0.6 is out 🚀 Prompt caching across Google/Anthropic/OpenAI/Amazon. If you’ve got repeated prompts in prod, this is a quick win on cost (and latency).

[Heinrich Wendel @hmw147]

🐒 Hebo Gateway 0.6

Prompt caching support across Google, Anthropic, OpenAI, and Amazon models.

Save 90% on your LLM calls by hosting your own gateway.

Release notes:
https://t.co/yiMmUENZMw

Got feedback or ideas for 0.7? Drop them in the repo! 🙌

#ai #llm #opensource

[Heinrich Wendel @hmw147]

One last drop for this week 🍌

🐒 Hebo Gateway v0.5:

📐 Structured Outputs (schema-first)
🎙️ Audio Inputs
💎 Gemini 3.1 Pro Preview
🧠 Claude Adaptive Thinking

Integrate the AI gateway directly into your app. 🚀

Runs on @honojs @elysiaJS @vercel @Cloudflare + anywhere else.

https://t.co/0lwZ8zs73G

[Heinrich Wendel @hmw147]

🐒 Hebo AI Gateway 0.4 just dropped — Observability built-in.

Self-host your AI gateway.
Connect @langfuse / OTEL in minutes.
Get traces + metrics + logs for your LLM traffic.

🚀 From zero → observable agents in <5 min.

https://t.co/2mzKt4qiSc