termaitrik 1.1.6

Terminal AI assistant inspired by Warp AI, supporting Ollama and BYOK

TermAI — AI assistant for your terminal (Ollama + BYOK)

Version: 1.1.6

TermAI is a terminal AI assistant (CLI + local server) inspired by Warp AI. It works offline with Ollama and supports BYOK (Bring Your Own Key) for cloud providers (OpenAI or compatible endpoints).

Features

• termai chat "..." — general chat (optional streaming)
• termai suggest "...description..." — generates shell commands with reasoning and risks
• termai explain --cmd "command" — explains what a command does
• termai fix --cmd "command" --error "stderr" — suggests a fix
• termai run "...description..." — suggests a command and asks for confirmation before executing it
• termai agent "...goal..." — experimental multi‑step iterative assistant (proposes & confirms each command)
• Local FastAPI server: uvicorn termai.server:app --host 127.0.0.1 --port 8765

Requirements

• Python 3.10+
• (Optional) Ollama at http://127.0.0.1:11434
• (Optional) cloud provider key for BYOK (OPENAI_API_KEY etc.)

Installation & Quick Start

Install from PyPI:

pip install termai
termai --help

Or run without installing:

uvx termai --help

For development (from source):

pip install -r requirements.txt
pip install -e .
termai --help

The shell integration alias (see below) provides resilient fallbacks that work across all install modes.

Configuration

Create ~/.termai/config.yaml (see examples/config.example.yaml).

Minimal example (Ollama):

default_provider: ollama
model: llama3.1:8b
ollama:
  host: http://127.0.0.1:11434

BYOK example (OpenAI):

default_provider: openai
model: gpt-4o-mini
openai:
  api_key: sk-...
  base_url: https://api.openai.com/v1

Useful environment variables: TERMAI_PROVIDER, TERMAI_MODEL, OPENAI_API_KEY, OPENAI_BASE_URL, OLLAMA_HOST.

Shell Integration (alias ai)

Install a resilient alias that falls back across install modes:

bash scripts/install-shell-integration.sh
# then reopen your shell (or: source ~/.bashrc | ~/.zshrc | fish config)
ai suggest "create a tar archive of the current folder"

Uninstall:

bash scripts/uninstall-shell-integration.sh

The alias resolution order:

1. Global/venv command termai
2. Importable module: python -m termai.cli
3. PyPI package via uvx termai
4. Local repository via uvx --from <repo> (development)

Local Server API

Start:

uvicorn termai.server:app --host 127.0.0.1 --port 8765

Endpoints:

• GET /health → { "ok": true }
• POST /v1/chat → body:

  {"messages":[{"role":"user","content":"hi"}],"provider":"ollama","model":"llama3.1:8b"}
  

  Response: { "content": "..." }

Safety & Warnings

• Always inspect and confirm suggested commands before executing.
• Redact secrets before sending errors or stack traces to providers.
• Local models (Ollama) keep data on your machine; cloud providers transmit prompts off‑device.

Changelog Highlights 1.0.0

• Added robust shell integration with multi-fallback launcher.
• Stabilized command set (chat, suggest, explain, fix, run, plus info/examples helpers).
• Improved provider error handling & streaming.
• Config merging & environment variable expansion.
• Extended test coverage across core flows.

MIT License.

Experimental Agent Mode

You can try an early iterative "agent" loop that plans and executes several shell commands with your confirmation between steps:

termai agent "list the latest 5 created files, then show the first"

Workflow per step:

1. Model emits JSON: { "thought", "command", "explanation", "done" }.
2. You confirm/modify/abort; if accepted the command runs locally.
3. Stdout/stderr are summarized and appended to the conversation as an observation.
4. Loop continues until done=true, command empty, or max steps reached (default 6).

Options (current minimal POC):

--steps N          maximum steps (default 6)
--dry-run          never execute commands (records hypothetical observations)
--model / -m       override configured model
--temperature / -t sampling temperature (default 0.1)

Roadmap ideas (not yet implemented): whitelist & yolo auto‑approval modes, danger pattern guard, transcript export, richer tool schema. Feedback welcome.