syntropylabs-evalkit 0.1.26

EvalKit Python SDK — LLM observability and tracing

EvalKit Python SDK

LLM observability and tracing for Python apps. One init() call auto-instruments your LLM clients, HTTP calls, database queries, and logging — then streams traces to Syntropy Labs.

Installation

pip install syntropylabs-evalkit

Optional provider extras:

pip install "syntropylabs-evalkit[openai]"      # OpenAI
pip install "syntropylabs-evalkit[anthropic]"   # Anthropic
pip install "syntropylabs-evalkit[all]"         # everything

The PyPI package is syntropylabs-evalkit, but you import it as evalkit.

Quickstart

import evalkit

evalkit.init(
    subscription_key="sk_...",       # your Syntropy Labs key
    service_name="my-service",
)

# That's it — your OpenAI / Anthropic / HTTP / DB calls are now traced automatically.
from openai import OpenAI

client = OpenAI()
resp = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Hello!"}],
)

init() sets up auto-instrumentation for you. Context (including trace IDs) propagates automatically across threads — no manual wiring required.

Web frameworks

# FastAPI / Starlette
from evalkit import EvalKitMiddleware
app.add_middleware(EvalKitMiddleware)

# Flask
import evalkit
evalkit.instrument_flask(app)

# Django — add to MIDDLEWARE
"evalkit.EvalKitDjangoMiddleware"

Manual spans

import evalkit

end, ctx = evalkit.start_span("my-operation", {"key": "value"})
try:
    ...  # your work
finally:
    end("ok")

Tracing your own functions & tools (APM)

Auto-instrumentation covers libraries (LLM / HTTP / DB). For your code, opt in with one decorator — a function, a tool, a whole class, or an entire module:

import evalkit

# One function -> function_call span (input / output / latency)
@evalkit.trace_function()
def do_work(x):
    return x * 2

# One tool -> tool_call span (renders in the Input/Output panels + tool metrics)
@evalkit.trace_tool()
def search_web(query: str):
    return run_search(query)

# Every method of a class, APM-style
@evalkit.traced
class OrderService:
    def place(self, order): ...
    def cancel(self, id): ...

# Every function defined in a module — one call
import myapp.services as svc
evalkit.trace_module(svc)

# Or your WHOLE app at once (recurses every submodule) — any framework:
import myapp
evalkit.trace_package(myapp)

Client-side tools you run yourself only show their output if you wrap them with trace_tool — the SDK sees the model's request but never your function's return value. Server-side tools (OpenAI web_search, …) and LangChain tools are captured automatically. Call init() before the decorated class/module is imported.

SQLAlchemy

import evalkit
evalkit.patch_sqlalchemy_engine(engine)

Evaluation

Score agent outputs locally — no judge-model cost, results appear as eval_result spans:

import evalkit

scores = evalkit.evaluate(
    output="Your return window is 30 days.",
    input="What is the return policy?",
    expected_tools=["search_knowledge_base"],
    tool_calls=[{"name": "search_knowledge_base"}],
    constraints={"required_terms": ["return", "30"]},
)
# → {"tool_trajectory_f1": 1.0, "required_terms": 1.0, ...}

Scenario simulation

Generate realistic synthetic-user scenarios from your agent's system prompt and tool list, then run each scenario against your real agent and score the results automatically:

import evalkit

evalkit.init(subscription_key="tk_live_...", service_name="my-agent")

# Step 1 — generate scenarios server-side (BYOK: your own key for the generation call)
scenarios = evalkit.generate_scenarios(
    agent_instructions=SYSTEM_PROMPT,
    tools=["search_kb", "lookup_order", "create_ticket"],
    count=5,
    provider="anthropic",           # "openai" or "google" also supported
    api_key="sk-ant-...",           # BYOK key for generation model
    model="claude-haiku-4-5-20251001",
)

# Step 2 — simulate each scenario against your real agent and score it
def entrypoint(ctx: evalkit.SimContext) -> evalkit.AgentTurnResult:
    # ctx.message    — the synthetic user's turn message
    # ctx.session_id — stable per-scenario, use it to keep multi-turn context
    reply, tools_used = run_my_agent(ctx.session_id, ctx.message)
    return evalkit.AgentTurnResult(
        text=reply,
        tool_calls=[{"name": t} for t in tools_used],
    )

report = evalkit.simulate_user(entrypoint, scenarios, tags=["ci"])
# Results appear in Dashboard → Simulations
print("Simulation ID:", report["simulation_id"])

Out-of-process agents (Claude Agent SDK)

The Claude Agent SDK runs the Anthropic call in a subprocess, so the normal in-process patch can't observe it. EvalKit wraps claude_agent_sdk.query() and ClaudeSDKClient.receive_response() instead, reading token/cost/latency from the ResultMessage the SDK already returns. This happens automatically via init() when claude_agent_sdk is installed. To call it explicitly:

evalkit.patch_claude_agent_sdk()

Flushing

Traces are batched and exported in the background. Flush before exit if needed:

evalkit.flush()

Links

• Website: https://syntropylabs.ai
• Documentation: https://syntropylabs.ai/docs

License

Proprietary — © 2026 Syntropy Labs. All rights reserved. See LICENSE.