Helicone

Model providers connect through the gateway, and Helicone also works with frameworks such as the Vercel AI SDK, LiteLLM, OpenRouter and CrewAI, sends webhooks out, and offers a Zapier app that runs chat completions through the gateway. These route model traffic and send events; none connects an agent to a real system to take actions.

SourceHelicone, docs.helicone.ai/llms.txt (integrations, webhooks, Zapier, web search)read 2026-09-21

Per request, the gateway routes and falls back between providers, and prompts are assembled from versioned templates at call time. No workflows that sequence, branch or retry agent steps, or combine deterministic nodes with agent steps, are documented.

SourceHelicone, docs.helicone.ai/llms.txt (provider routing, prompt assembly)read 2026-09-21

Model traffic is routed and logged, web search can be added to Anthropic models through the gateway, and datasets hold logged requests for evaluation and fine-tuning. No document ingestion, index, retrieval layer or knowledge API that grounds an agent in company data is documented.

SourceHelicone, docs.helicone.ai/llms.txt (web search, datasets)read 2026-09-21

With LLM Security enabled, Helicone screens each user message with Meta's Prompt Guard, and optionally Llama Guard across 14 threat categories, and blocks detected threats with an error response before the request is processed; moderation filters harmful content, and custom rate limits cap requests by count or cost.

SourceHelicone, docs.helicone.ai/features/advanced-usage/llm-security and llms.txt (moderations, custom rate limits); docs.helicone.ai/features/advanced-usage/llm-security.mdread 2026-09-21

The pricing page lists SAML SSO on the Enterprise plan, with identity integration through the customer's own provider and organizations and seats per plan. SOC 2 Type II and HIPAA compliance are listed from the Team plan up, with InfoSec reviews and customized MSAs on Enterprise; the data autonomy reference adds encrypted provider keys, EU or US regions and configurable retention. Roles or permissions inside an organization are not documented.

SourceHelicone, helicone.ai/pricing and docs.helicone.ai/references/data-autonomyread 2026-10-01

Every request through the gateway or SDK is logged with prompt, completion, tokens, cost, latency and errors, and Helicone groups them into sessions for agents and pipelines, adds user analytics and custom properties, and lets teams query analytics directly with HQL, a SQL interface with row-level security; data export is available, and retention runs 7 days on Hobby, 1 month on Pro, 3 months on Team and indefinitely on Enterprise, with configurable retention there.

SourceHelicone, docs.helicone.ai/llms.txt (sessions, HQL, data export) and helicone.ai/pricingread 2026-09-21

Logged requests, sessions and prompt versions persist, and the gateway can automatically clear old tool uses and thinking blocks from long-running agent sessions. No session, workflow or long term memory that an agent reads and writes is documented.

SourceHelicone, docs.helicone.ai/llms.txt (sessions, context editing)read 2026-09-21

Customers choose an EU or US region for data residency on Helicone Cloud, the open source platform self-hosts with Docker or Kubernetes and Helm, and the Enterprise plan lists on-prem deployment.

SourceHelicone, docs.helicone.ai/references/data-autonomy and llms.txt (self-hosting, Docker, Kubernetes), and helicone.ai/pricing; docs.helicone.ai/references/data-autonomy.mdread 2026-09-21

Helicone publishes a model registry and pricing data for the models its gateway reaches and cookbooks on building agents. No ready-made agents, templates or packaged workflows a buyer adopts are documented.

SourceHelicone, docs.helicone.ai/llms.txt (model registry, cookbooks)read 2026-09-21

Alerts fire when an application crosses error thresholds or cost limits, automated weekly reports deliver usage, cost and performance summaries to email or Slack, and webhooks send request events to the customer's endpoints. Each starts without a person asking, on a threshold or a schedule.

SourceHelicone, docs.helicone.ai/llms.txt (alerts, reports, webhooks) and helicone.ai/pricingread 2026-09-21

The Helicone AI Gateway gives one OpenAI-compatible API across more than 100 providers with automatic provider routing, fallbacks, caching, custom rate limits by request count or cost, and provider keys stored encrypted in a vault, so customers bring their own keys or use gateway credits.

SourceHelicone, docs.helicone.ai/llms.txt (AI Gateway overview, provider routing, error handling and fallback, custom rate limits) and docs.helicone.ai/references/data-autonomyread 2026-09-21

Helicone documents a REST API for requests, sessions, scores, evaluations, prompts, webhooks and trace logging, an OpenAI-compatible gateway API, SDK integrations, webhooks, a Zapier app, and a Helicone MCP server that lets MCP-compatible assistants query observability data; the codebase is open source.

SourceHelicone, docs.helicone.ai/llms.txt (REST reference, MCP server, webhooks, Zapier)read 2026-09-21

Scores and user feedback on requests and sessions come in through the API, and Helicone curates request data into datasets for fine-tuning and evaluation, provides a playground and versioned prompts deployed through the gateway, and tracks score distributions over time. Running an application against fixtures or datasets before production and configurable quality gates are not documented.

SourceHelicone, docs.helicone.ai/llms.txt (eval scores, datasets, prompts, evaluations API) and helicone.ai/pricingread 2026-09-21

Helicone documents observability, the AI Gateway, prompts, datasets, security features and self-hosting, and no browser, desktop or computer control by an agent is documented.

SourceHelicone, docs.helicone.ai/llms.txtread 2026-09-21