Files
Pizza 11b49d3933 feat: milestone 3 complete + QA benchmark
Web UI (tinyclaw web):
- Flask + SSE real-time progress streaming
- Queue view: add/delete tasks, status colours, error display
- Live log panel with coloured output
- History tab with unified diffs (Rosé Pine theme)
- Model picker dropdown
- Auto-refresh every 5s

History (history.py):
- JSONL-based, records before/after content + unified diff
- Wired into driver for EDIT and CREATE actions
- Failed tasks also recorded

QA benchmark (tinyclaw qa <models...>):
- 5 fixed tasks: divide/greet/is_even/repeat/safe_get
- Scores syntax validity + instruction following
- Comparison table: pass%, syntax%, latency, cost

QA results (3 models):
- gemma-3-12b-it:  80% pass, 100% syntax, ~944ms  ← winner
- gemma-3-4b-it:   60% pass, 100% syntax, ~917ms
- qwen2.5-coder-7b: 0% pass,   0% syntax (max_tokens issue — hits 512 limit mid-function)
2026-04-03 01:27:41 +00:00

194 lines
8.1 KiB
Python

"""TinyClaw CLI entry point."""
import argparse
import os
import sys
from tinyclaw.driver import RunConfig, run, status
def main():
parser = argparse.ArgumentParser(
description="TinyClaw: orchestrate tiny models for real project work",
formatter_class=argparse.RawDescriptionHelpFormatter,
epilog="""
Examples:
tinyclaw run --queue QUEUE.md --model qwen/qwen3.5-1.5b
tinyclaw run --dry-run
tinyclaw status
tinyclaw plan "add input validation to all API endpoints"
""",
)
sub = parser.add_subparsers(dest="command", metavar="command")
# ── run ──────────────────────────────────────────────────────────
run_p = sub.add_parser("run", help="Process task queue")
run_p.add_argument("--queue", default="QUEUE.md", help="Path to queue file (default: QUEUE.md)")
run_p.add_argument(
"--model",
default=os.environ.get("TINYCLAW_MODEL", "qwen/qwen3.5-1.5b"),
help="Model ID on OpenRouter or local endpoint (default: qwen/qwen3.5-1.5b)",
)
run_p.add_argument("--api-base", default=None, help="API base URL (default: OpenRouter)")
run_p.add_argument("--api-key", default=None, help="API key (default: OPENROUTER_API_KEY env)")
run_p.add_argument("--max-tokens", type=int, default=1024, help="Max tokens per completion")
run_p.add_argument("--temperature", type=float, default=0.1, help="Sampling temperature")
run_p.add_argument("--no-commit", action="store_true", help="Don't auto-commit after each task")
run_p.add_argument("--dry-run", action="store_true", help="Parse queue and show tasks without running")
run_p.add_argument("-v", "--verbose", action="store_true", help="Verbose output")
# ── status ───────────────────────────────────────────────────────
status_p = sub.add_parser("status", help="Show queue status")
status_p.add_argument("--queue", default="QUEUE.md", help="Path to queue file")
# ── web ──────────────────────────────────────────────────────────
web_p = sub.add_parser("web", help="Launch web UI")
web_p.add_argument("--queue", default="QUEUE.md", help="Queue file")
web_p.add_argument("--history", default="HISTORY.jsonl",help="History file")
web_p.add_argument("--host", default="127.0.0.1")
web_p.add_argument("--port", default=7842, type=int)
# ── qa ───────────────────────────────────────────────────────────
qa_p = sub.add_parser("qa", help="Benchmark small models")
qa_p.add_argument("models", nargs="+", help="Model IDs to benchmark")
qa_p.add_argument("--api-key", default=None)
qa_p.add_argument("-v", "--verbose", action="store_true")
# ── add ──────────────────────────────────────────────────────────
add_p = sub.add_parser("add", help="Add a task to the queue")
add_p.add_argument("action", choices=["EDIT", "CREATE", "DELETE", "TEST"], help="Task action")
add_p.add_argument("filepath", help="Target file")
add_p.add_argument("target", help="Function name, or - for whole file")
add_p.add_argument("instruction", help="What to do (quoted string)")
add_p.add_argument("--queue", default="QUEUE.md", help="Queue file to append to")
# ── plan ─────────────────────────────────────────────────────────
plan_p = sub.add_parser("plan", help="Decompose a feature into a task queue (uses big model)")
plan_p.add_argument("description", help="Feature description in plain English")
plan_p.add_argument(
"--model",
default=os.environ.get("TINYCLAW_PLAN_MODEL", "anthropic/claude-sonnet-4-5"),
help="Planning model (should be a big model)",
)
plan_p.add_argument("--queue", default="QUEUE.md", help="Output queue file")
plan_p.add_argument("--api-key", default=None, help="API key")
plan_p.add_argument("--context", default=None, help="Path to context file (e.g. ARCH.md)")
args = parser.parse_args()
if args.command is None:
parser.print_help()
sys.exit(1)
# ── dispatch ─────────────────────────────────────────────────────
if args.command == "run":
config = RunConfig(
queue_path=args.queue,
model=args.model,
api_base=args.api_base,
api_key=args.api_key,
max_tokens=args.max_tokens,
temperature=args.temperature,
dry_run=args.dry_run,
verbose=args.verbose,
auto_commit=not args.no_commit,
)
try:
run(config)
except FileNotFoundError as e:
print(f"Error: {e}", file=sys.stderr)
sys.exit(1)
except KeyboardInterrupt:
print("\nInterrupted.")
sys.exit(130)
elif args.command == "status":
print(status(args.queue))
elif args.command == "web":
from tinyclaw.webui import serve
serve(queue_path=args.queue, history_path=args.history, host=args.host, port=args.port)
elif args.command == "qa":
from tinyclaw.qa import run_qa
run_qa(args.models, api_key=args.api_key, verbose=args.verbose)
elif args.command == "add":
_run_add(args)
elif args.command == "plan":
_run_plan(args)
def _run_add(args) -> None:
"""Append a single task to QUEUE.md."""
from pathlib import Path
line = f'{args.action} {args.filepath} {args.target} "{args.instruction}"\n'
p = Path(args.queue)
if not p.exists():
p.write_text("# TinyClaw task queue\n")
with p.open("a") as fh:
fh.write(line)
print(f"Added to {args.queue}: {line.strip()}")
def _run_plan(args) -> None:
"""Call a big model to decompose a feature into QUEUE.md tasks."""
from pathlib import Path
from tinyclaw.client import generate
context = ""
if args.context and Path(args.context).exists():
context = Path(args.context).read_text()
system = """You are a senior software engineer breaking down feature requests into atomic code tasks.
Output ONLY a QUEUE.md task list in this exact format:
EDIT <file> <function_name> "<what to change>"
CREATE <file> - "<what to create>"
DELETE <file> <function_name> "<why>"
TEST <file> - "<pytest command or describe what to test>"
Rules:
- One task per line, no blank lines between tasks
- Each task touches exactly one function (or creates one new file)
- Instructions must be concrete and actionable, 10 words or less
- No explanations, no headers, just the task list"""
user = f"Feature: {args.description}"
if context:
user += f"\n\nProject context:\n{context[:2000]}"
print(f"Planning with {args.model}...")
print(f"Feature: {args.description}\n")
resp = generate(
prompt=user,
model=args.model,
system=system,
api_key=args.api_key,
max_tokens=2048,
temperature=0.3,
)
queue_path = Path(args.queue)
if queue_path.exists():
existing = queue_path.read_text().strip()
if existing:
print(f"Appending to existing {args.queue}")
queue_path.write_text(existing + "\n" + resp.content.strip() + "\n")
else:
queue_path.write_text(resp.content.strip() + "\n")
else:
queue_path.write_text(resp.content.strip() + "\n")
print(resp.content)
print(f"\nWritten to {args.queue}")
if resp.cost:
print(f"Cost: ${resp.cost:.4f}")
if __name__ == "__main__":
main()