Only to the LLM provider you choose (OpenAI, Google Gemini, or
Anthropic), and only the parts of the conversation needed to answer the
current turn. The Micracode core runs locally on 127.0.0.1 (in the
desktop app it runs in-process). There's no telemetry endpoint and no
Micracode–run cloud service.
No — one is enough. Configure whichever provider you want to use. The model picker only enables models from providers that have a key configured.
Out of the box, no. The core binds to 127.0.0.1 so it only accepts
connections from the same machine. Exposing it to a network would be a
significant change and would need real authentication, a tighter CORS
config, and an ingress in front. It's intended to run on your laptop,
for you.
On your filesystem, under ~/opener-apps/ (or wherever you pointed
OPENER_APPS_DIR). See Projects on Disk.
Yes. Each project folder is a normal repo. Install its dependencies
with bun install (or npm install) and run its standard scripts.
Yes — the model reads files fresh from disk each turn, so external edits are picked up the next time you chat. Just don't edit the same file in two places at once or you'll get a conflict on save.
Use the model picker in the chat composer at the bottom of the chat panel. The change applies to the next turn; your project's history and files are untouched.
Delete its folder under ~/opener-apps/. The home page recent-projects
list reads from disk, so the entry will disappear on next reload.
Yes. cd into the project folder and git init. The
.micracode/ subfolder contains the chat history (prompts.jsonl)
and project metadata (project.json) — commit them if you want to
preserve the conversation alongside the code, or .gitignore them if
you don't.
No. Micracode has no auth, no users, no accounts.
MIT — see the LICENSE file.