Zero-AI detection
Frequency data, acronym rules, and local context find technical terms without a model call. Code-shaped text is ignored.
Jargon radar for your agents
Keep your AI fast. Know the terms it uses.
unjargon finds technical language in Claude Code and Codex transcripts without spending AI credits, then gives every detected term a public reference.
Works beside your local CLI. No wrapper, no automatic AI calls.
live · claude-code · sim-pipeline
detected in this message
“The ODE system is stiff, so RK4 diverges. Switch to BDF.”
Original text stays verbatim; code paths and commands are ignored.
BDF · public referencestiff · public reference
Detection and public references are zero-AI. In-session explanations need confirmation.
A lighter kind of observability
Find jargon without slowing down your agent.
Public references first
Tap a term for a zero-AI Wikipedia summary plus Google and Wikipedia links. Ask for an in-session explanation only when you need it.
Works where you work
A tiny collector watches local Claude Code and Codex transcript files on your Mac, Linux box, or remote machine.
Made for real agent work
For the moments when “done” isn’t an explanation.
Vibe coding: understand a refactor, migration, or test failure before approving the next prompt.
Research & HPC: follow an agent’s simulation, data, or cluster work without becoming fluent in every acronym first.
Async collaboration: share a direct glossary of the terms an agent used, without asking collaborators to read an opaque transcript.
What stays private
Your raw transcripts, project paths, message annotations, and session-specific explanations stay scoped to your account. A public reference lookup sends only the detected term, never the surrounding transcript.
Sign in, pair a machine, then run the collector. Your AI-provider login does not give unjargon access to local files.
Get started in three steps
One sign-in. One pairing code. One command.
- 1 Sign in to unjargon with Google.
- 2 In your empty stream, choose Create pairing code.
- 3 Run this on the Mac or Linux machine where Claude Code or Codex runs. Paste the pairing code when prompted.
macOS · Linux
$ curl -fsSL https://raw.githubusercontent.com/Chrisa142857/unjargon.app/main/install.sh | sh -s -- --server https://unjargon.onrender.comNo root access. The installer downloads one static binary and registers a user-level service.