Klassiq AI — terminal coding agent

A fast, zero-dependency coding agent for your terminal, with the look and feel of modern agent CLIs. It talks directly to your AI provider (OpenAI-compatible /v1/chat/completions), so the model can read, search and edit files inside the folder you launch it in — always asking before it changes anything.

No API keys are hardcoded. Your key lives only in ~/.klassiq/config.json.

Install

npm install -g klassiq
cd anywhere && klassiq

Or from this repository: npm install -g . — and klassiq.cmd runs it in place on Windows.

Requires Node.js 18+.

First run

klassiq

You land straight in a full-screen, keyboard-driven TUI (like opencode). If no provider is configured yet it shows a hint — it never forces a wizard on you. Configure once with:

klassiq connect            # or press /connect inside the TUI

The wizard connects your provider (hidden API key entry), fetches your models, and lets you pick a default mode. Everything is stored in ~/.klassiq/config.json.

The TUI

• Type a message and press Enter to chat; the agent explores and edits on its own.
• @name — filter and pick a file to attach to the message (up/down + Enter).
• !command — run a shell command directly.
• / — open the command palette; keep typing to filter, Enter to run.
• ctrl+x then a key — quick actions: c compact · u undo · r redo · n new session · l sessions · m models · t themes · x export · q quit (matching the keys shown on screen).
• ? — toggle help.
• Sessions are saved automatically to ~/.klassiq/sessions/; klassiq -c resumes your last one, klassiq --session <id> opens a specific one.

Tool calls stream live. Nothing is ever applied silently:

• edit_file shows a colored diff and asks Apply? [y/N]
• bash shows the command and asks Allow? [y/N]
• every edit is snapshotted, so /undo and /redo revert it file by file
• plan mode blocks editing and commands entirely

Modes

Switch anytime with /mode (or /mode plan when you want analysis only).

| Mode | Tools | | ----- | ------------------------------------------------------------ | | build | read_file, list_files, grep_search, edit_file, bash | | plan | read_file, list_files, grep_search (read-only) | | chat | none (conversation only) |

Usage

• Pipe a question: echo "what does README.md say?" | klassiq
• One-shot: klassiq "question"
• Overrides: klassiq --provider openrouter --model <id> --key <key> "..."
• Env overrides: KLASSIQ_KEY, KLASSIQ_MODEL, KLASSIQ_PROVIDER.
• List models: klassiq --models

Commands

/connect                setup wizard (first-time setup)
/mode build|plan|chat   switch agent mode
/key                    change API key (hidden)
/model <id>             set the model directly
/models                 list models for this provider
/sessions               open a saved session
/new                    start a new session
/undo  /redo            revert / re-apply the last edited file
/compact                summarize the conversation and reset the context
/export                 write the conversation to a markdown file
/config                 show current config (key masked)
/clear                  reset conversation and attachments
/themes  /details  /thinking   UI toggles
/command                run a shell command directly (! also works)
/status                 read-only git status
/help                   show help
/quit                   exit

Safety

• Edits are confined to the folder where you launched klassiq.
• Nothing is written and no command runs without your y confirmation.
• Your files only ever go to the provider you configured, as tool results.
• Wipe settings anytime by deleting ~/.klassiq/config.json.

Providers

OpenAI, OpenRouter, TokenRouter, OpenCode, or any OpenAI-compatible host (custom, including local servers like Ollama/LM Studio via their /v1 URLs).

A note on custom/local hosts: model listing needs a working /models endpoint; if a provider is unreachable you can still type a model id manually during setup.