Skip to content

:hedgehog: PostHog is the leading platform for building self-driving products. Our developer tools – AI observability, analytics, session replay, flags, experiments, error tracking, logs, and more – capture all the context agents need to diagnose problems, uncover opportunities, and ship fixes. Steer it all from Slack, web, desktop, or the MCP.

2.6 KB
Raw Blame
import httpx
import openai
import anthropic
from prometheus_client import Counter

# All LLM API exceptions (for catching)
LLM_API_EXCEPTIONS = (anthropic.APIError, openai.APIError)

# Transient errors that may resolve on retry (5xx, rate limits, timeouts, connection issues)
LLM_TRANSIENT_EXCEPTIONS = (
    # Anthropic transient errors
    anthropic.InternalServerError,
    anthropic.RateLimitError,
    anthropic.APITimeoutError,
    anthropic.APIConnectionError,
    # OpenAI transient errors
    openai.InternalServerError,
    openai.RateLimitError,
    openai.APITimeoutError,
    openai.APIConnectionError,
)

# httpx transport errors that can escape SDK wrapping during streaming.
# These are network-level issues (not LLM provider errors) and are tracked
# on a separate counter to avoid false "LLM Provider Errors" alerts.
HTTPX_TRANSPORT_EXCEPTIONS = (httpx.ReadError, httpx.ConnectError)

# Client/validation errors that won't resolve on retry (400, 422, etc.)
LLM_CLIENT_EXCEPTIONS = (
    # Anthropic client errors
    anthropic.BadRequestError,
    anthropic.UnprocessableEntityError,
    # OpenAI client errors
    openai.BadRequestError,
    openai.UnprocessableEntityError,
)

LLM_PROVIDER_ERROR_COUNTER = Counter(
    "posthog_ai_llm_provider_errors_total",
    "Total number of LLM provider API errors (transient)",
    ["provider"],
)

LLM_CLIENT_ERROR_COUNTER = Counter(
    "posthog_ai_llm_client_errors_total",
    "Total number of LLM client/validation errors",
    ["provider"],
)

LLM_TRANSPORT_ERROR_COUNTER = Counter(
    "posthog_ai_llm_transport_errors_total",
    "Total number of httpx transport errors during LLM streaming",
    ["error_type"],
)

# Unexpected exceptions that crash an agent run (e.g. a code bug or a stale import),
# as opposed to the known LLM provider/client/transport errors counted above. These
# bubble to the catch-all in the runner and return a FailureMessage to the user — an
# agent generation that never finished. Sits at ~0 in steady state, so a sustained
# rate signals that a recent change broke an agent path in production.
AGENT_RUN_UNHANDLED_ERROR_COUNTER = Counter(
    "posthog_ai_agent_run_unhandled_errors_total",
    "Total number of agent runs that failed with an unexpected (non-LLM) exception",
    ["error_type"],
)


def resolve_llm_provider(exc: Exception) -> str:
    """Extract the LLM provider name from an exception.

    Only meaningful for SDK exceptions (anthropic.*, openai.*).
    """
    return type(exc).__module__.partition(".")[0] or "unknown_provider"


class GenerationCanceled(Exception):
    """Raised when generation is canceled."""

    pass