Arize Phoenix

Release 14.5.0 introduces new features for chat metadata, CLI annotations, and agent controls, alongside a fix for malformed pxi skill YAML files.

Version 14.4.0 introduces a new capability registry for agent tooling and resolves an issue by unpinning the graphql-core dependency.

Version 14.3.1 primarily focuses on a security enhancement by hardening the authentication mechanism during dataset uploads.

Version 14.3.0 introduces an assistant agent settings page and updates internal dependency handling for strawberry-graphql compatibility.

This patch release removes WebSocket support for GraphQL subscriptions, migrating to HTTP multipart for better compatibility and stability.

Version 14.2.0 introduces a new API endpoint for accessing projects by name and includes documentation updates.

Version 14.1.1 primarily addresses a bug related to connection exhaustion by implementing dataloader for session annotations.

Phoenix version 14.1.0 primarily focuses on bug fixes, including improvements to OpenAPI examples, agent error surfacing, and environment variable handling for SQL schema.

This patch release deprecates the evals 1.0 API and removes the associated legacy experiments module.

Version 3.0.0 introduces a major overhaul to the evals system by deprecating and removing the legacy evals 1.0 API and the associated experiments module.

Version 14.0.0 introduces major architectural changes, including a subcommand-first CLI structure and the removal of legacy evaluation and client plumbing. Key features include enhanced agent tooling, PostgreSQL read replica support, and ephemeral experiment handling.

This release introduces a new utility for converting ATiF traces into trajectory format.

Version 13.23.0 introduces bundled TypeScript SDK documentation in npm packages and resolves a bug with the Playground button in the prompts table.

Patch release 13.22.2 primarily addresses a database query optimization within the ProjectHasTracesDataLoader.

This patch release focuses on performance improvements by optimizing database queries and reducing redundant data fetching during initialization.

This release introduces new REST API endpoints for user information and secrets management, alongside adding the get_traces method to clients and supporting Python 3.14.

This release introduces official support for Python 3.14, with the exception of Windows environments.

This release introduces minor UI updates, including new icons, and fixes an installation issue specific to Windows and Python 3.14 environments.

This release introduces Python 3.14 support, enhances agent session summaries, and adds a REST endpoint for server secrets management.

This release introduces new REST API endpoints for managing prompt versions and tags, alongside UI improvements including a new CopyField component.

This patch release primarily addresses a usability bug in the UI related to model menu search focus.

This patch release includes a bug fix for TracesTable interface updates and documentation improvements for the phoenix CLI.

Version 13.19.0 introduces new UI components like PromptInput, enhances agent chat performance, and fixes a critical dependency issue with FastAPI.

This patch release primarily updates the dependency on arize-phoenix-evals to version 2.12.0.

Version 2.12.0 introduces support for structured data inputs in evals and patches several bugs, including a security-related pin on the LiteLLM dependency.

This patch release addresses a security concern by pinning the LiteLLM dependency to a safe version.

This release introduces a new prompt version diff view in the application and updates the rendering engine by replacing all markdown processing with streamdown.

Version 13.17.0 introduces an experimental simulated bash tool for agents and adds clipboard copying for spanIDs. This release also addresses several dependency vulnerabilities and improves compatibility with Starlette 1.0+.

This release introduces significant feature enhancements across skills, CLI, and UI, including new onboarding snippets and agent tooling. It also deprecates `useNotifyError` in favor of inline alert banners for error handling.

This release introduces new filtering capabilities for spans via the getSpans method in the client.

Version 2.0.1 primarily introduces a bug fix related to server version gating and capability requirements within the client.

This release introduces a new `phoenix db migrate` subcommand and enhances the server API with new trace retrieval and filtering capabilities. It also updates the internal client dependency and improves compatibility for certain LLM providers.

Version 2.0.0 introduces significant updates including new LLM provider integrations and the removal of the deprecated `client.annotations` in favor of `client.spans`.

This release introduces new features like the phoenix-pr-screenshot slash command and session conversation API support in clients. It also includes bug fixes related to dialog content scrolling.

Version 13.13.0 introduces UI filtering for model providers via environment variables and enhances the dataset view with drag-and-drop functionality. This release also removes the deprecated client.annotations module and splits side navigation APIs.

This release introduces new features like feature-flagged tracing onboarding and a new PackageManagerCommandBlock component, alongside several bug fixes across the playground, server, and UI components.

This release introduces major enhancements to the playground by adding built-in support for Perplexity AI and Together AI as new LLM providers, complete with model listings and cost tracking.

This release introduces four new first-class AI providers (Cerebras, Fireworks AI, Groq, Moonshot) with automatic cost tracking, updates the playground with the latest OpenAI models, and enhances the UI with editable project settings and improved experiment indicators.

Version 13.9.0 introduces filetype-agnostic dataset uploads and adds trace_id filtering to span retrieval endpoints. It also includes dependency updates and improvements to playground error handling.

This release introduces new session management capabilities to the Python client SDK, including session retrieval methods and enhanced session API options.

This release introduces new features like integration icons for Claude/LangGraph, Pydantic Vercel data streaming support for agents, and brute force login protection. It also includes several bug fixes and documentation updates for Claude Agent SDK integration.

This release updates the underlying container image to debian13 and patches dependencies by upgrading arize-phoenix-client and arize-phoenix-otel.

This release introduces expanded Python version compatibility for arize-phoenix-otel, supporting up to Python 3.14, and includes a minor fix to the Slack invitation link.

Version 1.30.0 introduces new GET endpoints for the sessions REST API, enhancing session management capabilities.

Version 13.6.0 introduces new UI annotation features and CLI support for multi-turn conversations, alongside several security and packaging bug fixes.

Version 13.5.0 introduces a configurable password policy and new GET endpoints for the sessions REST API, alongside dependency updates.

Version 2.11.0 introduces new evaluation capabilities to evals, including conciseness and refusal evaluators, alongside a fix for handling recent litellm version changes.

This release introduces several new features, including SQL visibility in stdout and navigation counters, alongside significant internal updates like migrating the application bundler to rolldown-vite and upgrading the AI SDK to v6. Numerous bug fixes address model name support, UI accessibility, and dependency updates.

This patch release (1.29.1) focuses on minor bug fixes, including updating documentation snippets and fixing an issue related to project deletion after experiment deletion.

Version 13.3.0 introduces a new conciseness classification evaluator and removes several UI components related to model inferences and embeddings. Bug fixes include improved error handling for JSONDistanceEvaluator.

Release 13.2.0 introduces support for AWS Bedrock cross-region inference preferences and includes several bug fixes, notably reverting a dependency that caused build issues.

Version 13.1.1 primarily addresses a bug fix related to the ModelMenu query fetch policy.

Version 13.1.0 introduces significant enhancements to table functionality, adds new session skills, and merges the phoenix-sdk-python skill into phoenix-evals. This release also includes various bug fixes related to model pricing and UI theming.

Patch release 13.0.3 primarily addresses a bug fix related to the playground interface.

Version 2.10.0 introduces dataset evaluators, full mustache prompt template support in the server, and enhancements to classification evaluators. A bug fix simplified template formatters.

This patch release primarily addresses dependency updates, specifically upgrading the arize-phoenix-client to version 1.29.0.

This release introduces dataset evaluators, enhancing the capabilities for evaluating datasets within the library.

Version 13.0.1 is a patch release primarily focused on fixing an issue related to file inclusion during package building.

Version 13.0.0 introduces extensive enhancements to LLM evaluation capabilities, including new built-in evaluators, improved playground features, and UI updates. This release contains breaking changes related to dataset evaluators.

This release introduces support for the Claude Opus 4.6 model and includes several bug fixes related to data validation, dependency updates, and improved UX for trace management.

Version 1.28.1 primarily addresses a bug by introducing timezone validation for the log_spans_dataframe function.

This release introduces the tool_selection evaluator to the evals library and includes several bug fixes, notably updating built-in model token prices and ensuring correct rendering of message content text.

This release introduces the new FaithfulnessEvaluator, several new evaluation metrics, and improves trace ID handling. A minor bug fix addresses the use of deprecated pandas NA checking.

Patch release 12.33.1 primarily addresses a bug fix related to GraphQL attribute access.

Version 12.33.0 introduces configurable email extraction via EMAIL_ATTRIBUTE_PATH for OAuth2 and resolves an import ordering issue related to namespace package discovery.

This release introduces several new skills, including CI/SKILL, evals, and tracing, alongside a new tool invocation accuracy metric. Bug fixes include updated token prices and improved dataset creation submission.

This patch release focuses on bug fixes, notably enabling multipart GraphQL subscription cancellation and removing the unnecessary pytz dependency.

This patch release updates the internal arize-phoenix-client dependency and fixes a KeyError issue related to pandas 3.0 compatibility in span queries.

This release introduces the new FaithfulnessEvaluator, deprecates HallucinationEvaluator, and adds support for linking dataset examples to traces via span_id_key.

This release introduces a new CLI, adds the FaithfulnessEvaluator, and includes updates to token prices and trace linking capabilities.

This release introduces new UI components for displaying JSON attributes and fixes several bugs related to preserving invocation parameters and response formats during model updates.

This release introduces connection timeout error handling, a correctness evaluator, and updates to span attribute display in the UI. It also includes dependency updates and documentation improvements.

Version 2.8.0 introduces a new correctness evaluator and enhances the LLM constructor by enabling sync/async client kwargs. It also fixes an issue related to newlines in built-in prompts.

Version 1.27.2 primarily addresses a bug related to using the context.span_id column when DataFrames utilize an integer index.

Patch release 12.28.1 primarily addresses a bug related to handling non-array values in NumDocuments to avoid PostgreSQL errors.

Version 12.28.0 introduces JSONL dataset upload support and a new tool selection correctness metric for evals, alongside several UI and layout bug fixes.

Version 12.27.0 introduces several UI/UX enhancements, including analytics, theme selection, and layout consistency, alongside a fix for built-in prompt formatting.

Version 12.26.0 introduces organizational improvements by moving the dataset example to the examples tab and includes several dependency updates and bug fixes, notably updating token prices and ensuring proper database engine disposal.

This patch release primarily addresses a bug fix related to updating the available options for the `reasoning_effort` setting when interacting with OpenAI services.

This patch release focuses on updating internal cost data and refining OpenAI reasoning effort options, alongside general documentation improvements.

Version 12.25.0 introduces lazy loading for embeddings/dimensions, Gemini tool call support, and updates key dependencies. LDAP configuration for email attributes has also been adjusted to use null as a sentinel value.

This patch release (2.7.1) primarily addresses bug fixes related to the rate limiter within the evals functionality.

Version 1.27.0 introduces new tracing capabilities with span notes functionality and adds support for Lightweight Directory Access Protocol (LDAP).

This release introduces optional scarf analytics and readme analytics features, alongside documentation updates to reorganize the sidebar navigation.

This release introduces a feature allowing the email field to be optional for LDAP configurations and includes documentation updates.

This release introduces a new span note function for the Python client and includes documentation updates.

Version 12.21.0 introduces a new endpoint for span notes and updates the built-in model token prices for cost tracking.

Version 12.20.0 introduces support for existing secrets and LDAP, alongside a dependency update for arize-phoenix-evals.

This release introduces support for prompt/template messages within the evals module.

Version 12.19.0 introduces a new feature for splitting on the experiments table and includes bug fixes related to token prices and Bedrock invocation parameters.

Version 12.18.0 introduces support for Claude Opus 4-5, enables split assignment during dataset upload, and includes several dependency updates and bug fixes.

Version 1.26.0 introduces the ability to assign splits directly during dataset uploads.

Version 1.25.0 introduces new evaluation helpers specifically designed to simplify pulling RAG spans.

Version 12.17.0 introduces new features like enabling repetitions in the playground and improving the retention policy UX, alongside several dependency updates and bug fixes across the platform.

Version 2.6.1 primarily addresses a bug related to handling None values for top_p and temperature settings within the AnthropicModel integration.

Version 1.24.0 primarily updates the client license to Apache 2.0.

This release primarily focuses on licensing, switching the client to the Apache 2.0 license.

This release introduces significant feature enhancements, including support for OpenAI 5.1 and expanded reasoning capabilities for Gemini models, while simultaneously dropping support for Python 3.9.

This patch release (12.15.1) focuses on maintenance, including updating built-in model token prices and fixing caching and integer handling issues.

Version 12.15.0 introduces an update to the supported Anthropic model list.