Your agent ran 15 steps. Step 9 failed silently. OpenJCK shows you exactly what happened — every LLM call, every tool, every token.
The dashboard has been completely redesigned with a new dark-mode architecture.
No config files. No setup. No SDK to learn. Wrap your entry point, your LLM calls, and your tools. OpenJCK does the rest.
from openjck import trace, trace_llm, trace_tool
@trace(name="my_agent")
def run(): ...
@trace_llm
def call_llm(messages): ...No wrapper scripts. No extra commands. python my_agent.py — and OpenJCK captures everything silently in the background.
$ python my_agent.py
[OpenJCK] Run complete → COMPLETED
[OpenJCK] 8 steps | 2840 tokens | 4.2s
[OpenJCK] → http://localhost:7823/trace/a3f9c1b2Open the trace viewer and see your entire agent run laid out step by step. Every LLM call, every tool call, every decision.
$ npx openjck
→ Opening OpenJCK at localhost:7823
● ── ● ── ● ── ● ── ● ── ● ── ● ── ✕
1 2 3 4 5 6 7 8The timeline highlights failures in red. Click any step. See the full input, output, error traceback, and token count.
STEP 8 — write_file [FAILED] 12ms
─────────────────────────────────────
ERROR
PermissionError: cannot write to output.md
File is open in another process
Line 47 in tools.pyYou know exactly what failed, why it failed, and what the agent was trying to do. Fix the bug with all the context you need.
# Before: file was locked by another process
# OpenJCK showed: output.md opened at step 3
@trace_tool
def write_file(path: str, content: str):
with open(path, 'w') as f: # ← fixed
f.write(content)Re-run your agent. Watch the timeline go green. Every trace is saved locally — compare runs and ship with confidence.
$ python my_agent.py
[OpenJCK] Run complete → COMPLETED
[OpenJCK] 9 steps | 2840 tokens | 3.8s
✓ All steps passedfrom openjck import trace
from my_app import run_agent
@trace(name="production_agent")
def main():
run_agent()