llm-evaltrack
Drop-in observability for LLM applications. Automatic quality scoring, hallucination detection, cost tracking, and agent run debugging — with 2 lines of code.
Install
pip install llm-evaltrack
Quick Start
import llm_observe
llm_observe.init(api_url="https://your-server.com/ingest")
llm_observe.patch_openai() # auto-track all OpenAI calls
llm_observe.patch_anthropic() # auto-track all Anthropic calls
That's it. Your existing code is unchanged. Every chat.completions.create() and messages.create() is now automatically tracked.
Agent Debugging (v0.2.0)
Trace multi-step agent runs to see every step, find where things break, and measure cost per span:
from llm_observe import trace_agent
with trace_agent("research_agent", input="Research renewable energy trends") as trace:
with trace.span("web_search", span_type="retrieval") as s:
results = search("renewable energy 2024")
s.set_output(results)
with trace.span("llm_summarize", span_type="llm", model="gpt-4o") as s:
summary = llm.summarize(results)
s.set_output(summary)
s.set_tokens(1200)
s.set_cost(0.009)
with trace.span("fact_check", span_type="tool") as s:
verified = fact_check(summary)
s.set_output(verified)
trace.set_output("Report complete")
Span types: llm · tool · retrieval · decision · custom
Each trace captures: total duration, tokens, cost, per-step timing, inputs/outputs, and errors.
Errors inside a with block are automatically caught and marked as failed.
What Gets Tracked (auto)
| Field | Source |
|---|---|
| Input / Output | Message content |
| Model | response.model |
| Tokens | response.usage |
| Cost (USD) | Calculated from token counts |
| Quality Score | Heuristic evaluation or LLM judge |
| Hallucination flags | Automatic detection |
Manual Tracking
llm_observe.track_llm_call(
input="What is the capital of France?",
output="Paris.",
prompt="You are a helpful assistant.",
model="gpt-4o",
metadata={"feature": "qa", "user_id": "u_123", "cost_usd": 0.0003},
)
Configuration
llm_observe.init(
api_url="https://your-server.com/ingest",
api_key="your-secret", # optional bearer token
max_retries=3,
timeout=5.0,
enabled=True, # set False in tests
)
Dashboard
Pair with the self-hosted server for:
- Real-time quality trend charts
- Bad response categorization
- Root-cause analysis by prompt
- Cost vs quality per model
- Regression alerts
- Agent Debugger — waterfall timeline of every span in a trace
Live demo: llm-evaltrack-production.up.railway.app
Self-host: github.com/Soufianeazz/llm-evaltrack
Changelog
v0.2.0
- Added
trace_agent()context manager for agent run tracing - Added
span()andtrace.span()for individual steps - Spans support:
set_output(),set_tokens(),set_cost(),set_error() - Automatic error capture on exceptions inside
withblocks - Nested span support via
parent_span_id
v0.1.0
- Initial release:
patch_openai(),patch_anthropic(),track_llm_call()
Metadata
Release files for llm-evaltrack 0.2.0
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| llm_evaltrack-0.2.0.tar.gz | 44.1 kB | Without provenance | Details for llm_evaltrack-0.2.0.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| llm_evaltrack-0.2.0-py3-none-any.whl | Python 3 | none | any | Details for llm_evaltrack-0.2.0-py3-none-any.whl |
Total release size: 54.1 kB
Release files / llm_evaltrack-0.2.0.tar.gz
| Download URL | llm_evaltrack-0.2.0.tar.gz |
|---|---|
| Size | 44.1 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
bb82b362476da790d995c59f2826476fa347a5e66d6886b9543cac4f7405fe56
|
|
BLAKE2b-256 checksum How to use checksums |
2085665ed2b77cf7a325d732106fe5f860e5b1c71287e13e565b02059ac246a9
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.2.0 CPython/3.12.10
|
Release files / llm_evaltrack-0.2.0-py3-none-any.whl
| Download URL | llm_evaltrack-0.2.0-py3-none-any.whl |
|---|---|
| Size | 10.0 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
a4cf04ddd4545a34d21a9a6172163878c8c7492dcfb259e0e35aa2bec7452adf
|
|
BLAKE2b-256 checksum How to use checksums |
9c788a8b337eb452f08c07afb3d732b357e3c0ebe921567a433c7c7faae601fe
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.2.0 CPython/3.12.10
|
Security for llm-evaltrack 0.2.0
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.