The terminal where AI agents meet real hardware.

Built for embedded developers who code with AI agents: agent alerts, resume after a restart, and your board one tab away. Free for every developer.

Works with Claude Code and Codex · v1.11.0 · Windows 10/11 and Linux · No sign-up

Claude Code or Codex fixes the driver on the left. The board reboots on the right. No window juggling.

Real Claude Code; Codex works the same. The board is simulated in the demo. Waiting time shortened.

With Agent serial access, the agent reads that boot log itself and asks you in the tab (Allow) before it types. How it works

Meet Neuro. It answers from your own datasheets.

Type @ in any tab and ask. A local model answers from the documents you imported and shows the section it used, on your PC.

  • Per board Keep each board's documents in its own Job Mode.
  • Right command Describe the task, get the command for the board you are on.
  • On your machine First run downloads Qwen2.5 Coder 1.5B Instruct (1.1 GB, 4 GB RAM). Free, with no limit on questions.
Neuro's sources: the runbook section 4.1 it used, with the matching words highlighted
Every answer comes with its source: the runbook section Neuro used, highlighted. Recorded offline.

NeuroFlash log search

Jev, cloud NeuroFlash log search finds the log lines behind a symptom: press Ctrl+Shift+F, switch to Ask and describe it. It runs in the cloud on your own OpenRouter account, shows the estimated cost, and sends nothing until you press Enter. How it works

Also in the box

Specifications

Connections
Serial, SSH with SFTP, WSL, PowerShell and cmd on Windows, your shell on Linux
Sessions
20 sessions at once, up to four panes per tab, one workspace per board
Agents
Claude Code and Codex CLI. Agent alerts, and Agent resume: after a relaunch each tab resumes its own Claude Code or Codex conversation. One hook install per host. Agent serial access (MCP): the agent reads your serial tabs and, when you allow it, types into them or borrows the port for flashing.
Neuro
Qwen2.5 Coder 1.5B Instruct, local. Free, with no limit on questions or documents.
Platforms
Windows 10/11 and Linux (.deb, .rpm, .AppImage)

What the app sends over the network

Console output is often under NDA. NeuroTerm itself sends nothing about your sessions. Local Neuro keeps your prompts and documents on your PC; the cloud options below send only what they say.

Rolling it out to a team? Who needs Commercial

Terminal output, prompts, documents with local NeuroStay on your PC
Telemetry and crash reportsNone
Update check GitHub, on by default, can be turned offAt launch
Model downloads Hugging Face; the embedding model downloads automatically when Settings opens with a local model readyWhen you pick one
NeuroFlash log search your symptom and the lines searched (selection, newest 2,000 lines or a chosen file), secrets redacted, to Jev on OpenRouterWhen you press Enter
Cloud AI you turn on prompts, terminal text, connection names, imported documents, secrets redacted; idle summaries, the journal and error explanations send automaticallyWhile on
Claude Code, Codex they use their own service, as in any terminal, including serial output they read through Agent serial access (MCP)Their provider

This is about the app. The website uses no analytics; see the privacy policy.

Free for you. Commercial seats for companies.

Commercial

$49 per user, per year

For companies of more than 10 employees that roll NeuroTerm out to their staff. The app never checks.