Opik

Integrations cover more than 40 AI frameworks, model providers and gateways, including LangChain, OpenAI, Google ADK, LangGraph and CrewAI, and alerts send webhooks out; these bring trace data in and send notifications, and none connects an agent to a business system to take actions. Through Opik Connect, Ollie reads the customer's agent codebase, writes approved fixes to it and reruns the agent from a failing trace. That is the one path that takes action, and it reaches only the local codebase.

SourceComet, comet.com/site/pricing and comet.com/docs/opik/llms.txt (Ollie, integrations)read 2026-09-21

From traces, Opik draws the customer's agent execution graphs, the Agent Playground runs the customer's agent locally while traced, and agent configuration versions prompts, model settings and tool definitions outside the codebase. No workflows that sequence, branch or retry steps, or combine deterministic nodes with agent steps, are documented as something Opik runs.

SourceComet, comet.com/docs/opik/llms.txt (development overview, agent playground) and comet.com/site/pricingread 2026-09-21

RAG metrics in Opik score retrieval quality; it also traces the context a customer's retrieval returns and generates synthetic examples from traces for datasets. No document ingestion, index, retrieval layer or knowledge API that grounds an agent in company data is documented.

SourceComet, comet.com/docs/opik/llms.txt and comet.com/site/pricingread 2026-09-21

Guardrails in Opik run inline with LLM calls on live traffic and can stop or alter a response before it reaches a user, with five guard types (PII, allowed and restricted topics, prompt injection and jailbreak, an LLM-as-a-judge check, and a custom classifier) grouped into named policies that can be enforced workspace-wide, and every run logged as a span. Ollie, Opik's assistant, proposes code edits that a person reviews and approves before anything changes on disk.

SourceComet, comet.com/docs/opik/guardrails/overview and comet.com/site/pricing (Ollie, Opik Connect); comet.com/docs/opik/guardrails/overview.mdread 2026-09-21

The Enterprise plan lists enterprise SSO over OAuth 2.0, SAML and LDAP with SSO enforcement, custom role-based access at project and organization level, view-only users, service accounts, and SOC 2, ISO 27001, ISO 9001, HIPAA and GDPR compliance; the documentation covers SAML, OIDC and JWT authentication and workspace roles and permissions, and the product page says Opik produces audit logs for governance teams.

SourceComet, comet.com/site/pricing, comet.com/docs/opik/llms.txt (authentication, roles and permissions) and comet.com/site/products/opikread 2026-09-21

Opik traces every step of an application's execution, from context retrieval to model responses and tool calls, with agent execution graphs, sessions for multi-turn conversations, token and cost tracking and error surfacing; traces, spans, threads, datasets and experiments export through the SDKs, REST API, UI and command line, span retention is 60 days on Free and Pro and custom on Enterprise, and the product page says Opik produces audit logs for governance teams.

SourceComet, comet.com/site/pricing, comet.com/docs/opik/llms.txt (observability, export data) and comet.com/site/products/opikread 2026-09-21

The customer's traces, sessions and threads are recorded, datasets and prompts are versioned, and an offline fallback replays tracing messages after an outage. No session, workflow or long-term memory that an agent reads and writes is documented.

SourceComet, comet.com/docs/opik/llms.txt (tracing concepts, offline fallback) and comet.com/site/pricingread 2026-09-21

Opik is open source and can be downloaded, installed and run on the customer's own infrastructure from the same codebase as the hosted versions; Opik Cloud stores data in the US on the Free and Pro plans, and Enterprise offers flexible deployments across cloud, on-premises and fully managed options with a custom data region.

SourceComet, comet.com/site/pricingread 2026-09-21

Opik ships Ollie, a ready-made assistant that reads traces, searches the workspace, builds test suites, proposes fixes and, through Opik Connect, edits and reruns the customer's agent; the same assistant powers Diagnostics, which scans traces for recurring issues. It is one packaged agent with preset uses, not a set of ready-made agents or templates a buyer selects among.

SourceComet, comet.com/docs/opik/llms.txt (Ollie, diagnostics) and comet.com/site/pricingread 2026-09-21

Online evaluation rules score production traces automatically as they are logged, Diagnostics scans a project's traces with Ollie and surfaces recurring agent issues with a root cause and suggested fix, and alerts send webhook notifications on events such as trace errors, new feedback scores and prompt changes. Evaluation and investigation start on incoming traces without a person asking.

SourceComet, comet.com/docs/opik/llms.txt (online evaluation rules, diagnostics, alerts) and comet.com/site/pricingread 2026-09-21

Through its Gateway, Opik gives a consistent API across multiple LLM providers, workspace AI Providers settings configure connections to providers, built-in LLM-as-a-judge metrics can run on a custom model, the Agent Optimizer works with OpenAI, Anthropic, Gemini, Azure and Ollama, and agent configuration keeps model settings outside the codebase, versioned and changeable without redeploying. Model choice stays with the customer, using its own providers.

SourceComet, comet.com/docs/opik/llms.txt (Gateway, AI Providers, custom model, configuring LLM providers) and comet.com/site/pricing (agent configuration)read 2026-09-21

Opik publishes a REST API with an OpenAPI specification and a complete Python client that work against both the open source platform and Opik Cloud, Python and TypeScript SDKs, native OpenTelemetry ingestion for other languages, the opik command line, and an MCP server that lets an AI coding assistant instrument code, query traces and manage prompts.

SourceComet, comet.com/docs/opik/llms.txt and comet.com/site/pricingread 2026-09-21

Test suites run pass or fail assertions at item and suite level with execution policies for how many runs must pass, experiments test an application over datasets with more than 30 built-in and custom LLM-as-a-judge, heuristic and code metrics for single steps, whole agents and full conversations, annotation queues gather expert review, online evaluation scores production traces, and the Agent Optimizer tunes prompts, tools and parameters with several algorithms.

SourceComet, comet.com/site/pricing and comet.com/docs/opik/llms.txt (test suites, online evaluation, Agent Optimizer)read 2026-09-21

The documentation and plan comparison cover tracing, evaluation, optimization, guardrails and administration, and no browser, desktop or computer control by an agent is documented; the Agent Playground runs the customer's own agent locally with tracing.

SourceComet, comet.com/site/pricing and comet.com/docs/opik/llms.txtread 2026-09-21