Co-Founder | CEO at https://nitter.cf/t.co/m3MAo3fMIZ Founder at Lattivox Labs. Building @arcdotfun @ryzomeai
Joined July 2015
- Tweets1.1K
- Following648
- Followers13K
- Likes5.5K
Pinned Tweet
AI models get most of the attention. But the model is only one part of an agent.
Rig just crossed two million downloads. I wrote about the infrastructure around the model, what reliable agents require, and why this less visible layer matters.
we build rig, rig builds us
really excited for you @gold_electrum, i can’t wait to tune in
I will be doing a livestream about Rust+AI and Rig with @orhundev , with support from @rust_foundation and @jetbrains . Please join us August 27 @ 16:00 UTC.
rustfoundation.org/media/jet…
youtube.com/live/hhw741JrQeQ
Tachi retweeted
Rig 0.36.0 is out! The AI library for all Rustaceans.
New providers: MiMo, MiniMax, Moonshot, Copilot, Z.AI
New models: Opus 4.7, GPT 5.5, GPT Image 2
Using your ChatGPT or Copilot subscription instead of API.
Full changelog: github.com/0xPlaygrounds/rig…
we build rig, rig builds us! looking forward to many more milestones and projects built with rig !
Rig has surpassed 7k stars on @github ! Giant THANK YOU to everyone who has contributed to and used Rig!
github.com/0xPlaygrounds/rig
Rig 0.35 is live!
this release includes:
tool calling that works across providers including llama.cpp, streaming that preserves turn history correctly, and dynamic model discovery.
we also shipped GPT Image 1.5, Bedrock OTel + prompt caching, and 100+ live regression tests.
change logs:
github.com/0xPlaygrounds/rig…
Rig 0.35.0 is out! The premier AI library written in Rust.
We added support for llama.cpp and GPT image 1.5, implemented dynamic model listing, improved response streaming and tool execution, and added 100+ live regression tests.
full changelog:
github.com/0xPlaygrounds/rig…
Tachi retweeted
rig v0.33 has been released!
- new McpClientHandler to make it easier to use MCP servers
- you can now use system messages directly rather than only thru preamble
- additional helpful tracing telemetry
check it out: github.com/0xPlaygrounds/rig…
looking for a CT intern.
specs:
- you post consistently and actually give a shit
- you understand crypto twitter culture natively
- community management comes naturally to you
- you're online when it matters
you don't need to be told what's funny, what's cringe, or what hits on CT. you already know.
reply with your twitter + why you.
Tachi retweeted
rig is... rust crate of the year???
watch @CodeToTheMoons talk about the number 1 AI framework in rust: youtube.com/watch?v=9L9oOmqD…
Tachi retweeted
iT's jUsT FinTecH CollEcTivE tHaT nOw oWnS ~5% moving $aixbt
how about no, total holders are absolutely exploding
the recent ai agent run is a reminder that @aixbt_agent is still our most useful agent today for anything crypto data or trading related
we are exploring code mode in rig.
this is not about adding JavaScript execution.
it is about changing how agents operate.
today an agent reasons in tokens.
it describes logic in natural language.
it calls a tool.
it continues reasoning.
that approach wastes tokens and blurs computation with narration.
with code mode the model writes executable logic.
the runtime evaluates it.
the result flows back into the agent.
the shift is simple.
from LLM as narrator to LLM as planner and compiler.
we believe this matters.
LLMs are weak at multi step math, state tracking, deterministic logic, and repetitive transformation.
letting them externalize these tasks into code reduces hallucination surface area.
it makes behavior reproducible.
it makes execution inspectable.
it also separates concerns.
the LLM plans.
the runtime executes deterministically.
tools expose capability.
as agents scale, context windows expand and costs compound.
execution environments scale better than token loops.
most frameworks wrap tool calls.
few provide embedded runtimes with developer defined APIs bridged cleanly to Rust.
code mode moves rig closer to an agent runtime framework.
closer to an operating system for agents.
we are designing the architecture now and would value serious feedback.
we build rig, rig builds us
introducing... code mode on rig!
we're adding functionality in rig for js code to be natively executed by a runtime instead of standard JSON tool calling.
this aims to provide lower token use and more efficient context windows over time through letting the model execute js code directly (as well as, of course, higher levels of effectiveness due to how much code there is in LLM training datasets)
let us know if you have any thoughts!
github.com/0xPlaygrounds/rig…
Tachi retweeted
there are so many cool new crypto x ai experiments happening onchain it's once again hard to keep up.
why this run could end differently than the last crypto x ai run of late 2024?
simple. autonomous agents are actually building real product, and finally earning $ for their token holders.
ai agents are new economic actors, and permissionless blockchains are proving to be their preferred substrate for interacting with the world.
117 contributors. 21+ releases. One lead dev who changed the slope of our entire curve. A year of building infrastructure for the people building AI. This one's as much about the people as the code.
we build rig. rig builds us.
Tachi retweeted
12 Days of Ryzome - Day 8 is for focusing on connections.
In your canvases you can build connections between ideas and research with arrows, allowing the nodes to link and connect. You can then group and colour nodes to further organise your thoughts. Your eyes process color before you consciously read anything, so a complex canvas suddenly becomes navigable in a way it wasn't before.
This mimics the way our brains work much more than traditional linear AI planning software, and sparks new areas of creativity that may not have been obvious without these connections
Discover this for yourself - ryzome.ai/
(for the best experience, use desktop 💻)