First run — Earshot docs

First run#

List what you can reach:

earshot models

anthropic/claude-opus-5              1000k       $5/$25  reasoning vision
anthropic/claude-sonnet-5            1000k       $2/$10  reasoning vision
openai/gpt-5.2                        400k    $1.75/$14  reasoning vision
google/gemini-flash-lite-latest      1049k    $0.3/$2.5  reasoning vision
...
896 models across 18 providers (catalog 2026-09-03)
prices are USD per million tokens, input/output

Filter by provider, model id or name:

earshot models opus
earshot models groq

Run a single turn:

earshot -p "explain what a wire adapter is in this codebase"

Pick a model explicitly — the default is anthropic/claude-opus-5:

earshot -p "hello" --model openrouter/anthropic/claude-opus-5
earshot -p "hello" --model ollama/qwen3-coder

Get structured output instead of prose:

earshot -p "hello" --output-format json

{
  "text": "Hello.",
  "model": "anthropic/claude-opus-5",
  "usage": { "inputTokens": 9, "outputTokens": 3 },
  "costUsd": 0.00012
}

-p uses the same agent loop and tools as the interactive UI. Permission requests cannot be answered interactively in headless mode, so select an appropriate permission mode or configure explicit rules for automation.

Work interactively#

cd your-project
earshot

Start with a bounded request such as “Explain how authentication works here.” Use /model to inspect or switch the active model, /permissions to inspect trust, and /help for commands. Type while the agent works to queue steering; Escape interrupts.

Resume work#

earshot --continue
earshot --resume

Resuming appends to the existing history. Permission grants are not restored from the transcript.