techrevati-runtime¶

Production-aware runtime primitives for multi-step LLM agent loops.

Beta — 0.1.x

The 0.x API surface is unstable; minor breaking changes are possible until 0.2.0. See Migrating from 0.0.x for the current upgrade notes, and the Changelog for what changed in each release.

pip install techrevati-runtime

What's in the box¶

Orchestrator — one session wires lifecycle, retry classification, circuit breaker, permissions, cost tracking, and policy
Retry policy — typed failure scenarios with recipe-driven recovery steps
Circuit breaker — three-state (CLOSED / OPEN / HALF_OPEN) protection
Usage tracking — per-model cost with caller-provided pricing
Quality gate — graduated pass/fail evaluation
Agent lifecycle — validated state machine with audit log
Agent events — typed lifecycle events with OpenTelemetry attribute bridge
Permissions — deny-first role × tool gating
Policy engine — composable rules over a phase context

Design tenets¶

Zero runtime dependencies — stdlib only
Type-safe (PEP 561 py.typed, clean under mypy --strict)
Thread-safe primitives
Configuration-free at the edges — caller supplies pricing, thresholds, roles

Quick example¶

from techrevati.runtime import (
    Orchestrator, UsageSnapshot, CircuitBreaker,
    register_pricing, ModelPricing,
)

register_pricing("model-a", ModelPricing(input_per_million=3.0, output_per_million=15.0))

orch = Orchestrator(
    role="writer",
    phase="draft",
    circuit_breaker=CircuitBreaker("model-api", failure_threshold=5),
)

with orch.session() as session:
    result, usage = session.run_turn(
        lambda: call_model(prompt),
        model="model-a",
        usage=UsageSnapshot(input_tokens=5000, output_tokens=1200),
    )

print(session.summary())

Next steps¶

Getting Started — installation and full walk-through
Primitives — module-by-module reference
Changelog