Fast, one-shot LLM queries for your terminal.
brew tap bwl/ettio
brew install karl
karl initkarl run "your question or task"
karl run "analyze auth.go for security issues"
karl route plan --json "implement the verifier"
karl route plan --json "read-only assess this repo; do not edit files"
karl tourcat error.log | karl run "what went wrong?"
git diff | karl run "write a commit message"karl models add argon --provider openrouter --model google/gemini-4-argon
karl run --model argon --no-tools --reasoning-effort high \
--max-tokens 262144 --timeout 2h --context-file context.txt \
"Analyze all the supplied material and write a substantive, evidence-based report." \
> answer.mdContext files and stdin are read in full. Argon's long responses can span
multiple provider requests; Karl follows its continuation tokens within the
chosen output budget. See long-context runs for the
reusable deep-read profile, background runs, and limits.
Named configurations that become verbs:
karl stacks create review # Create a stack
karl review "check auth.ts" # Use it as a commandkarl run <task> # Run a task (default stack)
karl route # Plan/select a brokered run route
karl init # First-time setup
karl providers # Manage LLM providers
karl models # Manage models
karl stacks # Manage config stacks
karl skills # Manage agent skills
karl tour # Meet Karl (local; no model call)- Product Contract
- Repository Workflow
- Product Direction
- Work Tracking
- Architecture Decisions
- Implementation Plans
- CLI Reference
- Configuration
- Tools
- TUI
- Volley Mode
- Architecture
- July 2026 Consolidation Handoff (historical)
- Run Architecture Roadmap (exploratory)
- Agent Skills
bun install
bun run buildMIT