Articles, notes and updates on the agent, the tools, and the craft of building software with AI.
A chatbot answers and stops. An agent loops — observe, reason, act, observe again — until the job is genuinely done. Strip away the features and that quiet cycle is all Elyra really is.
Conductor 0.7.5 turns the timeline into a one-click bridge to Elyra and the container badge into a doorway — context bundles out, container actions in, and the host still never reasons.
Twelve small, low-effort habits that change how Elyra feels to use — set a goal and walk away, rewind without fear, switch models mid-thought, and let the tool carry more than you usually hand it.
No framework, no build step, no orchestration spaghetti. In Elyra, an agent is just a folder and a command. Start from an empty directory and end with a working agent that has a brain, abilities, and memory of its own.
Nemotron 3 Ultra is open, strikingly cheap, and carries a million-token context window — and Elyra can drive it today. What it is, where it shines, where it doesn't, and the hardware story behind it.
MiniMax M3 lands on the direct minimax and minimax-cn providers — 512k context, 128k output, reasoning, and vision — plus an Anthropic SDK bump and a boring-but-honest release-tooling fix.
Elyra's @elyracode/semantic-index brings local semantic code search to the agent — find code by what it does, not what it's called, without your source ever leaving the machine.
The model that wrote your code is the worst reviewer of it. Elyra's /review gets you a cold, independent second opinion — no conversation history, optionally a different model entirely, scoped to exactly the changes you care about.
Elyra 0.9.3 ships extension_write — the agent can now author its own extensions, with your sign-off.
Elyra's /bisect borrows git's binary-search debugging and points it at the agent's own turns — finding the exact one that introduced a regression.