sillm

AI Cost Tracking

• 🤖 LLM usage: $1.2644 (20 commits)
• 👤 Human dev: ~$356 (3.6h @ $100/h, 30min dedup)

Generated on 2026-06-03 using openrouter/qwen/qwen3-coder-next

Shell LLM control plane for the semcod/coru ecosystem.

SILLM owns the shell-client side of LLM automation: clients such as aider, claude, codex, gemini, qwen-code, opencode, and devin. GUI/IDE chat control stays in koruide and the existing Koru autopilot socket path.

Commands

sillm clients
sillm drive --client aider --prompt "Refactor ticket PLF-1"          # dry-run
sillm drive --client aider --prompt "Refactor ticket PLF-1" --execute
sillm nlp "aider: napraw testy dla kolejki"                          # NLP -> DSL
sillm validate                                                       # ecosystem hooks + intents

sillm drive always saves the prompt under .koru/sillm/prompts/ before it executes or prints a dry-run plan.

Client notes:

• aider
• claude-code
• aider Docker autoloop

Optional ecosystem integrations

PyPI wheels ship only the dev extra. Monorepo siblings use local editable installs (PyPI rejects file:// URLs in package metadata):

pip install -e ".[dev]"
pip install -e ../nlp2dsl ../intract ../redsl ../proxym ../llx   # adjust paths for your checkout

The nlp2dsl bridge is opt-in at runtime:

export SILLM_NLP2DSL=1
export NLP2DSL_BACKEND_URL=http://localhost:8010
sillm nlp "uruchom claude dla ticketu PLF-123"

Without the service, SILLM falls back to a deterministic local parser.

sillm validate reports optional package availability and exposes the @intract.v1 intent contract used to validate sillm.drive DSL before shell execution.

License

Licensed under Apache-2.0.