# Agent Ping

> Production observability for AI agents. Cost attribution, live monitoring, quality scoring. One ingest pipeline, one dashboard, three answers about every agent you ship: what it costs, whether it's running, whether its answers are still good.

## Main Pages

- [Homepage](https://agentping.io/): AI agent observability platform overview
- [Features](https://agentping.io/features): Deep dive on Spend, Pulse, and Verify
- [Spend](https://agentping.io/features/spend): AI agent cost monitoring and LLM cost tracking, attributed per agent, customer, feature and run
- [Pulse](https://agentping.io/features/pulse): AI agent heartbeat and uptime monitoring, run-level traces, alerts to Slack, PagerDuty or webhook
- [Verify](https://agentping.io/features/verify): AI agent quality monitoring with deterministic checks plus LLM-as-judge evaluations and drift detection
- [Pricing](https://agentping.io/pricing): Free plan plus three flat tiers ($99 / $199 / $399). Named limits, no metered billing, unlimited seats and teams on every plan
- [Pricing (machine-readable)](https://agentping.io/pricing.md): Full tier breakdown, annual prices, event caps, and overage rules as parseable markdown
- [LLM API pricing](https://agentping.io/llm-api-pricing): Per-million-token prices for every model from OpenAI, Anthropic, Google Gemini, xAI, Mistral, DeepSeek, Cohere and TypeSafe (Jev), verified against each provider's page; the default rate card Agent Ping prices runs with
- [LLM API pricing (machine-readable)](https://agentping.io/llm-api-pricing.md): The same table as markdown; JSON at https://agentping.io/llm-api-pricing.json
- [Claude API pricing](https://agentping.io/llm-api-pricing/claude): Every Anthropic model per million tokens with cached input, cache write, batch, worked monthly bills and caching maths
- [OpenAI API pricing](https://agentping.io/llm-api-pricing/openai): Every GPT model per million tokens with cached input, batch, the 272k long-prompt surcharge and worked monthly bills
- [Gemini API pricing](https://agentping.io/llm-api-pricing/gemini): Every Google Gemini model per million tokens with cached input, batch, the 200k surcharge, scheduled price changes and worked monthly bills
- [Token counter](https://agentping.io/token-counter): Free token counter for Claude (exact via Anthropic's count_tokens), OpenAI GPT (exact, o200k_base in the browser) and Gemini (estimate), with the cost of the text on each current model
- [Glossary](https://agentping.io/glossary): Plain-English definitions for every term Agent Ping uses
- [What is AI agent observability?](https://agentping.io/what-is-ai-agent-observability): Definition, the three failure modes it solves, and a minimal implementation
- [Documentation](https://agentping.io/docs): SDK guides, API reference, product concepts. Every docs page is also served as raw markdown; append .md to its URL (e.g. https://agentping.io/docs/quickstart.md)
- [Blog](https://agentping.io/blog): Notes on running AI agents in production
- [Register](https://agentping.io/register): 14-day free trial, no card required
- [MCP server](https://agentping.io/docs/guides/mcp): Connect ChatGPT, Claude, Claude Code, Cursor, VS Code, Codex or Gemini CLI to your Agent Ping account

## Connecting an AI assistant

Agent Ping has a remote MCP server at `https://agentping.io/mcp` (Streamable HTTP, OAuth 2.1 sign-in, no keys to paste). Once connected, an assistant can list teams and agents, search runs, read run events and evaluations, and on paid plans acknowledge incidents, give evaluation feedback and change monitoring configuration. Every change is recorded in the Activity log. Setup for each client is at https://agentping.io/docs/guides/mcp.

## Capabilities

Agent Ping answers three questions about your production AI agents:
- **Spend**: What is each agent costing per run, per customer, per feature?
- **Pulse**: Is anything wrong right now? Live monitoring plus schedule freshness for cron-style agents.
- **Verify**: Is quality still holding? Deterministic checks plus LLM-as-judge scoring on a sampled subset.

## Solutions (who Agent Ping is for)

By team:
- [For SaaS teams](https://agentping.io/solutions/saas): Attribute AI cost per customer and feature, protect margins, catch quality drift before users do
- [For automation agencies](https://agentping.io/solutions/agencies): Monitor client AI workflows after handover, catch silent failures, bill AI cost back per client
- [For product teams](https://agentping.io/solutions/product-teams): Track AI feature quality in production, catch drift after a prompt change, tie cost and reliability to features
- [For platform and AI engineering](https://agentping.io/solutions/platform-engineering): Production observability for fleets of agents, incidents, schedule freshness, cost guardrails

By industry (teams most likely running agents in production blind):
- [Customer support](https://agentping.io/solutions/customer-support): Score AI deflection and triage bot answers, catch off-policy replies, track cost at support volume
- [Sales and RevOps](https://agentping.io/solutions/sales): Track AI SDR and enrichment cost at scale, score message quality, catch sequences that stop running
- [E-commerce and retail](https://agentping.io/solutions/ecommerce): Cost of catalog enrichment per SKU, product copy quality, overnight catalog jobs that fail
- [Legal and professional services](https://agentping.io/solutions/legal): Score accuracy of AI research and drafting, catch hallucinated citations, cost per matter

## Guides (how to monitor X, by framework)

- [Monitor LangGraph agents](https://agentping.io/guides/monitor-langgraph-agents): attach AgentPingCallbackHandler, capture cost/status/quality per graph run
- [Monitor OpenAI Agents SDK agents](https://agentping.io/guides/monitor-openai-agents-sdk): pass AgentPingHooks, capture hand-offs and tool calls per run
- [Monitor CrewAI agents](https://agentping.io/guides/monitor-crewai-agents): instrument_crewai(), one record per crew kickoff
- [Monitor Pydantic AI agents](https://agentping.io/guides/monitor-pydantic-ai-agents): instrument_pydantic_ai(), structured-output quality
- [Monitor LangChain agents](https://agentping.io/guides/monitor-langchain-agents): AgentPingCallbackHandler for chains and agents
- [Monitor LlamaIndex agents](https://agentping.io/guides/monitor-llamaindex-agents): AgentPingLlamaIndexHandler, RAG query cost and answer quality
- [Monitor AutoGen agents](https://agentping.io/guides/monitor-autogen-agents): instrument the provider and wrap the conversation in a run
- [Monitor Semantic Kernel agents](https://agentping.io/guides/monitor-semantic-kernel-agents): provider instrumentation or OpenTelemetry
- [Monitor Vercel AI SDK agents](https://agentping.io/guides/monitor-vercel-ai-sdk-agents): wrap generate/stream with the TypeScript SDK
- [Monitor Mastra agents](https://agentping.io/guides/monitor-mastra-agents): TypeScript SDK wrap or OpenTelemetry
- [Track LLM cost per customer](https://agentping.io/guides/how-to-track-llm-cost-per-customer): tag runs and attribute spend per customer

## Learn (problem-first pages)

- [Why is my OpenAI bill so high?](https://agentping.io/token-bill-too-high): Causes of a runaway OpenAI or Anthropic bill (retry loops, prompt creep, model swaps, long sessions), finding which agent caused a spike, per-agent cost attribution
- [Your AI agent stopped running, nobody got paged](https://agentping.io/silent-ai-agent): Why APM can't see absence-of-signal, schedule freshness with a tolerance window, alert payload anatomy
- [Your LLM output is getting worse, the dashboard says green](https://agentping.io/llm-quality-drift): How quality drift hides, deterministic checks plus LLM-as-judge, drift detection on the score distribution

## Compare

- [Agent Ping vs LangSmith](https://agentping.io/vs/langsmith): Framework-agnostic agents, per-customer attribution, schedule freshness, live drift detection on the production stream
- [Agent Ping vs Helicone](https://agentping.io/vs/helicone): SDK over proxy, agent run as the unit of analysis, quality scoring with calibration anchors
- [Agent Ping vs Langfuse](https://agentping.io/vs/langfuse): Hosted vs self-hostable, agent run vs trace, schedule freshness and anomaly detection on per-agent spend
- [Agent Ping vs Braintrust](https://agentping.io/vs/braintrust): Production observability vs offline evaluation; cost attribution and schedule freshness in addition to quality scoring
- [Agent Ping vs Datadog](https://agentping.io/vs/datadog): Purpose-built for agents, no custom-metrics homework, schedule freshness on AI agents, LLM-as-judge included
- [Langfuse vs LangSmith](https://agentping.io/compare/langfuse-vs-langsmith): Neutral head-to-head on pricing model, free tier, self-hosting, retention, framework fit and evals, with a worked bill at a million runs
- [LangSmith alternatives](https://agentping.io/compare/langsmith-alternatives): Nine tools compared on billing unit, hosting and best fit, with verified prices; Agent Ping is listed and labelled as ours
- [Langfuse alternatives](https://agentping.io/compare/langfuse-alternatives): Nine tools compared on billing unit, hosting and best fit, with verified prices; Agent Ping is listed and labelled as ours

## SDKs

- [Python SDK](https://agentping.io/docs/sdks/python): install from github.com/agent-ping/agent-ping-python. Python 3.10+, native auto-instrumentation for Anthropic and OpenAI, non-blocking background delivery.
- [TypeScript SDK](https://agentping.io/docs/sdks/typescript): install from github.com/agent-ping/agent-ping-typescript, imports as @agentping/sdk. ESM, Node 18+, Proxy-wrapped client instrumentation, AsyncIterable streaming wrap.
- [Go SDK](https://agentping.io/docs/sdks/go): github.com/agent-ping/agent-ping-go. Stdlib-only, zero external dependencies, http.RoundTripper-based instrumentation.
- [Laravel SDK](https://agentping.io/docs/sdks/laravel): composer require agentping/laravel. Auto-discovers and listens to laravel/ai events for zero-config observability.

## API Reference

- [Authentication](https://agentping.io/docs/api/auth): API keys and per-agent ping tokens
- [Runs](https://agentping.io/docs/api/runs): POST /v1/runs, append events, finish run
- [Events](https://agentping.io/docs/api/events): llm_call, log, tool_call payload shapes
- [Heartbeats](https://agentping.io/docs/api/heartbeats): One-shot run for webhook and cron integrations
- [GET /v1/ping](https://agentping.io/docs/api/ping): URL-shaped heartbeat for shell scripts
- [Rate limits](https://agentping.io/docs/api/rate-limits): Per-tier sustained rates and monthly event caps
- [Errors](https://agentping.io/docs/api/errors): Status codes, Retry-After semantics, SDK retry behaviour

## Integrations (no SDK required)

- [cron](https://agentping.io/docs/integrations/cron): wget or curl heartbeats from crontab
- [curl](https://agentping.io/docs/integrations/curl): Smoke-test and ad-hoc recipes
- [GitHub Actions](https://agentping.io/docs/integrations/github-actions): Workflow heartbeats with success or failure status
- [n8n](https://agentping.io/docs/integrations/n8n): HTTP Request node setup
- [Zapier](https://agentping.io/docs/integrations/zapier): Webhooks by Zapier action

## Product docs

- [Spend](https://agentping.io/docs/spend): Cost attribution per agent, customer, and feature, priced server-side from a rate card
- [Budgets](https://agentping.io/docs/spend/budgets): Monthly caps per agent or team, alerts at the thresholds you set
- [Spend guard](https://agentping.io/docs/spend/guard): Opt-in top-of-script gate that refuses to start a run once a spend rule trips; Team and Business plans
- [Spend alerts](https://agentping.io/docs/spend/alerts): Baseline anomaly detection and monthly budgets with threshold alerts
- [Pulse](https://agentping.io/docs/pulse): Live monitoring, schedule freshness, and alerting for agents in production
- [Heartbeats](https://agentping.io/docs/pulse/heartbeats): One-shot run reporting from cron, curl, or any webhook
- [Heartbeat freshness](https://agentping.io/docs/pulse/freshness): Expected schedules, the overdue sweep, and recovery notices
- [Status pages](https://agentping.io/docs/pulse/status-pages): Public, link-shareable agent health pages for clients and teams; shows states, never data
- [Verify](https://agentping.io/docs/verify): Quality scoring on every run, from deterministic checks to LLM-as-judge
- [Verify alerts](https://agentping.io/docs/verify/alerts): Check failures, rubric score thresholds, fail streaks, quality drift
- [Alert channels](https://agentping.io/docs/alerts/channels): Slack, Microsoft Teams, PagerDuty, email, and signed webhooks; delivery and retry behaviour

## Billing

- [Tiers](https://agentping.io/docs/billing/tiers): Free ($0), Starter ($99), Team ($199), Business ($399), Enterprise
- [Founding cohort](https://agentping.io/pricing.md): First 50 paying accounts lock founding pricing for life
- [Free trial](https://agentping.io/docs/billing/trial): 14 days full access, no card required
- [Quotas and rate limits](https://agentping.io/docs/billing/quotas): Per-second rate plus monthly event allowance, soft cap at 100% and hard cap at 150%

## Blog (Selected)

- [Three blind spots when running AI agents in production](https://agentping.io/blog/three-blind-spots-running-ai-agents-production)
- [What is AI cost attribution?](https://agentping.io/blog/what-is-ai-cost-attribution)
- [Anatomy of an £8,900 token bill](https://agentping.io/blog/anatomy-of-an-8900-token-bill)
- [How to catch a silent AI agent before your customer does](https://agentping.io/blog/catch-silent-agent-before-customer-does)
- [Quality drift: the bug you only find in your support inbox](https://agentping.io/blog/quality-drift-bug-support-inbox)
- [Beyond Sentry and Datadog: what observability for AI agents actually means](https://agentping.io/blog/beyond-sentry-datadog-agent-observability)

## Company

- Website: https://agentping.io
- Category: SaaS, AI Agent Observability, LLMOps, Cost Attribution, Quality Monitoring