SynapseKit v1.2.0 — Deploy, Track Costs, Evaluate

SynapseKit v1.2.0 ships three capabilities that every production AI team needs but rarely gets from a framework: one-command deployment, cost intelligence, and built-in evaluation.

synapsekit serve — Deploy in one command

pip install synapsekit[serve]
synapsekit serve my_app:rag --port 8000 --reload

Point it at any Python object that's a RAG pipeline, graph workflow, or agent — SynapseKit auto-detects the type and builds a FastAPI app with the right endpoints (/query, /run, /stream, /health). OpenAPI docs included. No boilerplate.

CostTracker & BudgetGuard

Cost control without a SaaS subscription:

from synapsekit import CostTracker, BudgetGuard, BudgetLimit

tracker = CostTracker()
guard = BudgetGuard(BudgetLimit(daily=10.0, per_request=0.50))

with tracker.scope("pipeline"):
    rec = tracker.record("gpt-4o", input_tokens=1000, output_tokens=500)
    guard.check_before(rec.cost_usd)

print(tracker.summary())

BudgetGuard uses a circuit breaker pattern — when a budget is exceeded, it trips OPEN and stops requests until the cooldown passes.

@eval_case + synapsekit test

@eval_case(min_score=0.8, max_cost_usd=0.05)
def test_summarization():
    result = run_my_pipeline("Summarize this document...")
    return {"score": evaluate(result), "cost_usd": tracker.total_cost_usd}

synapsekit test tests/ --threshold 0.7 --format json

Integrates naturally with pytest and CI. Exit code 1 if any case fails its thresholds.

What else is new

• PromptHub — local versioned prompt registry
• PluginRegistry — community plugins via Python entry points
• RedisCheckpointer + PostgresCheckpointer — production graph persistence

Full changelog: v1.2.0