opsai

AI terminal command helper. Ask in plain English, get the exact command — no long explanations, no chatter.

$ opsai 'roll back the last deployment in kubernetes'
  > kubectl rollout undo deployment/my-app
    Reverts the deployment to its previous revision
    Details
      Use `kubectl rollout history deployment/my-app` to list all
      revisions, then `undo --to-revision=<n>` to pick a specific one.

Forgot that one kubectl, docker, git or systemd incantation? Working with Kubernetes, EKS, or plain shell and can't recall the flag? Just ask opsai in plain English and copy the answer straight into your terminal.

Best use case: fully local with Ollama

opsai shines as a 100% local, private, offline-friendly setup using Ollama with a tiny model. A small 0.5B model is perfect for this — command completion is a simple task, so the small model answers fast on CPU with no GPU, and your queries never leave your machine.

ollama pull qwen2.5:0.5b
opsai --base-url http://localhost:11434/v1 --model qwen2.5:0.5b 'get the pods in the kube-system namespace'

Want it as the default? Save it once:

opsai --config
# Base URL: http://localhost:11434/v1
# Model:    qwen2.5:0.5b
# API key:  (not asked - local servers need no key)

Bigger local models (7B+) give even better answers on capable hardware — the interface is identical.

Install

pip install opsai

Works with any OpenAI-compatible API

OpenAI, Groq, Ollama, LM Studio — anything speaking the /v1/chat/completions protocol:

Usage

opsai 'reboot into bios'              # one-shot
opsai --chat                          # interactive session with memory
opsai --model gpt-4o-mini 'list pods' # per-invocation override
opsai 'kubectl restart deployment'    # k8s, git, docker, anything

More examples

Only your question is sent to the model — the table above is what the answer looks like. Always review the command before running it.

Per-invocation overrides: --base-url, --model, --api-key. Persistent (non-secret) settings live in ~/.config/opsai/config.json. Precedence: --flag > env var > config file > defaults.

API keys (when not using local models)

opsai --config stores your key in the OS keyring (GNOME Keyring / KWallet / Keychain / Credential Manager) — never written to disk. Local servers (Ollama/LM Studio) need no API key at all — opsai detects a localhost base URL and skips the key requirement. No keyring backend? Use the environment:

export OPSAI_API_KEY="your-key"      # or OPENAI_API_KEY
export OPSAI_BASE_URL="https://api.groq.com/openai/v1"   # optional
export OPSAI_MODEL="llama-3.3-70b-versatile"             # optional

Security

• Never executes anything. opsai only displays commands; there is no code path that runs them.
• Sends only your query plus a fixed system prompt — no shell history, no environment, no telemetry.
• API key lives in the OS keyring or an env var. Never logged, redacted from all error output, never sent on HTTP redirects.
• HTTPS only (except localhost for local model servers).
• Danger flagging: rm -rf, sudo, mkfs, dd, curl | sh, etc. are printed with a warning banner.
• Terminal escape sanitization: model output is scrubbed of ANSI/control characters before display.

License

MIT