tell-ai

One-shot terminal assistant.

What It Includes

• tell — terminal CLI for one prompt at a time

Usage

npm install -g tell-ai

Run a prompt:

tell "explain this directory"
tell d "run ls -la"
tell -m d "run ls -la"
tell -m --help

Command execution is interactive by default:

tell d "run ls -la"       # asks before executing
tell -y d "run ls -la"    # executes without confirmation
tell --no-exec d "run ls" # never executes requested commands

-y/--yes is intended for disposable or sandboxed environments. The CLI blocks known high-risk command patterns, but this is a heuristic guard, not a security boundary. Do not use automatic execution in production, critical hosts, or trusted workstations unless it is contained by a real sandbox such as a container or VM.

Persistent context across sessions:

tell -c "remember that this project uses PostgreSQL"
tell -c "now add a users table migration"   # remembers the previous message

Context is stored per working directory and model under ~/.ai/tell_context. Without -c, each invocation starts fresh.

Multi-step chain mode — the assistant can run a command, see its output, and continue with follow-up commands until it reaches a final answer:

tell --chain "find out why the build is failing and fix it"

Include piped input with a prompt:

npm run build 2>&1 | tell -i "what should I fix first?"
git diff | tell --input "review this change"

Tell logs conversations under ~/.ai/tell_history.

API Keys

API keys go in ~/.config/<vendor>.token, for example:

~/.config/openai.token
~/.config/anthropic.token
~/.config/google.token
~/.config/xai.token
~/.config/fireworks.token

Security Tests

Run the prompt-injection and command-execution safety checks with:

npm run test:security

License

MIT2