reeprov-client

Official Python client SDK for Reeprov — local-first debugging and reproducibility infrastructure for AI agents.

Installation

pip install reeprov-client

Requires Python >=3.10. Zero third-party dependencies required (built strictly using standard library urllib).

Quick Start

import os
from reeprov_client import ReeprovClient

# Initialize client
reeprov = ReeprovClient(
    url=os.environ.get("REEPROV_URL", "http://127.0.0.1:3000"),
    ingest_key=os.environ.get("REEPROV_INGEST_KEY", "ree_live_..."),
    project_id="customer-support",
    workflow_id="triage-v1",
)

# 1. Start execution run
run_id = reeprov.start_run(summary="Triage customer inquiry #5812")

# 2. Record model calls & token usage
reeprov.model_call_completed(
    run_id=run_id,
    model="claude-sonnet-4-6",
    provider="anthropic",
    prompt="Classify urgency: System outage reported.",
    response={"category": "P0_OUTAGE", "escalate": True},
    input_tokens=150,
    output_tokens=32,
    cost_usd=0.0008,
)

# 3. Record tool invocations
reeprov.tool_call_completed(
    run_id=run_id,
    tool_name="pagerduty.trigger_incident",
    arguments={"severity": "critical", "service": "payments"},
    output={"incident_id": "INC-9901"},
    permission_class="external_side_effect",
)

# 4. Complete the run
reeprov.complete_run(run_id=run_id, summary="Urgent outage escalated successfully.")

Environment Variables

The client can automatically configure itself from standard environment variables (with legacy REEPROV_* fallbacks during migration):

Reliability & Failure Isolation

• Zero Third-Party Dependencies: Pure Python standard library implementation (urllib.request).
• Fail-Open Posture: By default, telemetry delivery failures print non-fatal warnings to stderr without crashing the host agent.
• Fast-Fail on Auth: 401/403 responses fail immediately without wasteful retries.
• Bounded Exponential Backoff: Transient 5xx server errors retry up to 2 times.

License

Apache-2.0