iTerm2 for Agentic CLI Work

Short answer: in iTerm2, place the cursor in your agentic CLI prompt (Claude Code, Codex, Gemini CLI, OpenCode, Kimi CLI, etc.), press ⌃+⌥+R, speak, then press ⌃+⌥+R again. AICHE inserts clean text at the cursor.

Primary Use Case: Agentic AI CLI Prompts

iTerm2 is now the home for agentic dev workflows, not just shell commands. The typing bottleneck is no longer ls or cd. It is prompt quality:

• precise task scope and constraints
• guardrails and non-goals
• test and verification instructions
• file boundaries and repo context
• rollback and risk notes

One vague line gets vague edits. A 30-second dictated prompt gets high-quality output.

Agentic CLI Tools In iTerm2

This workflow applies directly to:

• Claude Code
• Codex CLI
• OpenCode
• Gemini CLI (see also Antigravity CLI for Google's 2026 consumer migration)
• Kimi Code CLI
• similar prompt-driven terminal agents

AICHE inserts text only. The agent tool executes planning, edits, and command runs.

App-Native iTerm2 Scenarios

Split panes for multi-agent orchestration

A common iTerm2 setup:

• pane 1: primary agent session
• pane 2: secondary agent/session for parallel task
• pane 3: test runner or logs

Dictate a full prompt into pane 1:
"Refactor auth middleware in src/server/auth.ts and src/server/session.ts. Keep public API unchanged. Add tests for expired and malformed tokens. Do not touch UI files. Return a short change summary and risk list."

Then dictate a validation prompt into pane 2:
"Review pane 1 changes for regression risk in token refresh flow and cookie flags. Focus on SameSite and secure defaults. Suggest only minimal diffs."

You get parallel high-context execution without prompt fatigue.

SSH + remote agent sessions

Many teams run agentic CLIs on remote devboxes over SSH from iTerm2. Dictate environment-aware prompts:
"On remote host, inspect failing CI reproduction in services/payments. Do not install new dependencies. Run existing test script only. If failure is flaky, show two candidate stabilizations and pick lowest-risk option."

This keeps remote constraints explicit.

Commit and PR messaging after agent changes

After agent edits land, dictate accurate commit/PR text while diff is open:
"Fix refresh-token race in session renewal by serializing renewal per user session. Adds regression tests for concurrent refresh and stale token eviction. No schema changes."

This preserves human-readable history after AI-assisted changes.

Legacy terminal prose still matters (secondary)

A common iTerm2 setup is three panes: logs, command runner, and notes. Keep pane 1 tailing logs, pane 2 for commands, and pane 3 in vim INCIDENT.md. Dictate timestamped notes in pane 3 while debugging in pane 2:
"00:42 UTC - restarted worker pool, queue lag dropped from 14m to 3m. 00:47 - DB connection spikes continue, likely retry storm from API gateway."

That gives you a usable incident timeline without pausing command work.

tmux handoff notes for shift changes

Inside tmux, teams often leave minimal context before detach. Dictate a proper handoff in a notes.md buffer:
"Migrations complete on staging only. Production blocked by index lock timeout on orders table. Next safe command is migration 024 rollback, not full reset."

The next engineer can resume safely.

Safer commit messages in terminal editors

For one-line subjects, inline entry is fine. For anything with rationale, run git commit and dictate subject + body inside the editor:

• subject: what changed
• body paragraph 1: why this approach
• body paragraph 2: risk/test notes

This avoids shell quoting issues from git commit -m and preserves proper history.

SSH session journaling on remote hosts

When connected to production via SSH, create a quick SESSION_NOTES.md and dictate what you changed:
"Applied nginx config for upstream timeout 45s. Reloaded successfully. Did not restart app processes. Verify 504 rate after 10 minutes."

That documentation often matters more than the command transcript.

iTerm2 Features That Pair Well With Dictation

• Split panes and pane titles. Name panes by agent role so prompts go to the right target.
• Broadcast Input (careful use). Keep dictation in one focused pane only; do not broadcast prompts accidentally to multiple shells.
• Profiles and badge names. Use profile/badge to confirm you are in the correct environment before dictating into remote editors.
• Instant Replay. Combine replay with dictated notes to reconstruct incident sequences faster.
• Triggers and annotations. Dictate the explanation, then add iTerm2 annotations for key log lines.

How It Works

1. Focus the exact iTerm2 pane and agent prompt input.
2. Press ⌃+⌥+R to start recording.
3. Speak complete scope, constraints, and expected output format.
4. Press ⌃+⌥+R again to insert.
5. Review prompt once, then run.

Audio is streamed for cloud transcription, processed, and discarded immediately after processing, within 1 second. No persistent audio copy.

FAQ

Need a plugin in iTerm2?
No.

Works inside agentic CLI prompt fields?
Yes.

Works inside vim insert mode and nano too?
Yes.

Does this work in tmux and SSH sessions?
Yes, as long as the correct pane/session has focus.

Offline?
Recordings queue and process when back online.

Try it now: open iTerm2 with two panes, run your agentic CLI in pane 1, then dictate a full prompt with file scope, non-goals, and test requirements. Compare output quality versus your usual short typed prompt.