v1.1.0 Release

Highlights

• Added support for glm 4.6 and kimi-k2 (including extended thinking mode).
• Introduced TokenEvent for RL and a new ConversationErrorEvent.
• Improved visualization of CondensationRequest
• Telemetry improvements: automatic log directory creation
• Observability: ad support for OpenTelemetry via Laminar.
• FileEditor now supports reading image file.

What's Changed

Added

• Support glm 4.6 by @simonrosenberg in #1046
• Support kimi-k2-thinking by @ryanhoangt in #1093
• Add image file viewing support to FileEditor by @SmartManoj in #1016
• Support kimi-k2 extended thinking, fix prompt caching stats, fix max output by @xingyaoww in #1133
• examples: add observability example by @enyst in #1131
• add TokenEvent and TokenEvent processing for RL by @lintangsutawika in #1123
• visualizer: render CondensationRequest nicely by @enyst in #1113
• Auto-create log directory for telemetry by @SmartManoj in #1111

Fixes and Improvements

• Fix Datadog log-query to use Logs API instead of Error Tracking API by @neubig in #1043
• fix: recursively parse nested tools params by @Edward-sy in #1011
• refactor: git SHA-based Docker tags by default, versioned tags only on releases by @xingyaoww in #1088
• fix race during cleanup by @enyst in #1095
• Revert "Add FallbackRouter for LLM failover support" by @xingyaoww in #1105
• Fix Remote Runtime API Authentication by @xingyaoww in #1090
• We now have an explicit ConversationErrorEvent in the stream by @tofarr in #1106
• Remove default path from ENV_ROOT assignment by @SmartManoj in #1050
• Use shell:bash in run-examples workflow by @xingyaoww in #1057
• Allow override git ref/sha in docker build by configuring BuildConfig by @xingyaoww in #1100
• Fix: LLM telemetry logging by @enyst in #1028
• fix(condenser): When condensation is triggered by the unhandled_condensation_request condition, it will result in empty condensation. by @CLFutureX in #1034
• Enable Ruff checks for mutable defaults (B006, B008, B039, RUF012) by @enyst in #1055
• Handle nested field schema formatting for tool description by @ryanhoangt in #1001
• Fix resume reconciliation by treating litellm_extra_body as runtime override by @enyst in #1115
• tests(llm): add explicit retry behavior tests for empty choices by @enyst in #1107
• SDK: export ConversationExecutionStatus via public API by @enyst in #1126
• fix(backend): unable to create a new conversation using the planning agent’s tools. by @hieptl in #1125
• Simplify API Workspace example with image_pull_policy argument; fix erroneous clean up of client by @xingyaoww in #1127
• refactor: move TokenEvent from llm_convertible to event by @xingyaoww in #1130
• Fix success rate calculation to exclude skipped tests by @xingyaoww in #1136
• Refactor: Always include risk fields by @malhotra5 in #1052

New Contributors

• @Edward-sy made their first contribution in #1011
• @lintangsutawika made their first contribution in #1123

Full Changelog: 1.0.0...v1.1.0

v1.0.0 Release

Official V1 SDK Release 🎉

• Documentation: https://docs.openhands.dev/
• Evaluation: https://docs.google.com/spreadsheets/d/1wOUdFCMyY6Nt0AIqF705KN4JKOWgeI4wUGUP60krXXs/edit?gid=811504672#gid=811504672
• Paper: https://arxiv.org/abs/2511.03690

What's Changed

• Agent Server UUIDs without hyphens by @tofarr in #960
• Enhance README with branding and resources by @xingyaoww in #961
• Default to function calling enabled, remove FUNCTION_CALLING_PATTERNS allowlist by @xingyaoww in #956
• Rename SecretsManager to SecretRegistry by @enyst in #965
• Rename ConversationState field to secret_registry with read-alias by @enyst in #969
• Add codex-mini-latest to RESPONSES_API_PATTERNS by @YakshithK in #934
• Fix double dollar sign in visualizer metrics display by @xingyaoww in #938
• fix browser tool cleanup by @dinmukhamedm in #942
• feat: Add llm_response_id to all LLM generated events by @ApGa in #930
• Fix formatting: remove extra blank line in model_features.py by @xingyaoww in #973
• Fix VNC security vulnerability by defaulting to disabled by @rbren in #974
• Fix wrong server path for file upload endpoint by @ryanhoangt in #951
• Fix bash executor session reuse by resetting closed sessions by @softpudding in #964
• Fix duplicate reviewer assignment in assign-reviews workflow by @neubig in #827
• Add logging for file upload and download actions by @rbren in #993
• sdk(llm): Stop model-name whack-a-mole: revert to core family substring matching by @enyst in #879
• Add glm4.6 to FORCE_STRING_SERIALIZER by @simonrosenberg in #994
• update patterns in FORCE_STRING_SERIALIZER_PATTERNS by @simonrosenberg in #995
• Add tests for string serializer by @enyst in #996
• Add cost reporting for run-examples.yml and schedule scripts to run daily by @ryanhoangt in #977
• Remove default MCP servers from preset agents by @xingyaoww in #984
• Bug: fix Visualizer "Message To User" that should be "Message from User" by @simonrosenberg in #1002
• Tweak review workflow for draft PRs by @enyst in #997
• Fix(llm):When LLM calls are retried, it can lead to inaccurate records. by @CLFutureX in #949
• Normalize SDK errors for clients by @enyst in #980
• Add Datadog debugging example with API integration tools by @neubig in #272
• Add Sphinx-compatible docstrings to core SDK classes by @rbren in #1006
• Refactor ToolDefinition architecture to use subclass pattern for all tools by @xingyaoww in #971
• Fix datadog debugging workflow example by @neubig in #1009
• feat(backend): enhance search_conversation_events functions with timestamp filtering support by @hieptl in #695
• Add comprehensive unit tests for conversation_router.py endpoints by @tofarr in #1010
• Update example API key in README by @SmartManoj in #1013
• Migrate to Blacksmith official Docker actions with native multi-platform builds by @xingyaoww in #990
• Fix/remove docstring for OpenHands tool package by @rbren in #1018
• feat(llm): forward extra_headers to LiteLLM completion and responses by @enyst in #733
• refactor: rename AgentExecutionStatus -> ConversationExecutionStatus & state's agent_status -> executiong_status by @simonrosenberg in #839
• Add OpenTelemetry with Laminar SDK by @dinmukhamedm in #681
• More agent_status to execution_status_updates, and error state by @tofarr in #1019
• Recognize GPT‑5 context overflow and trigger condenser (Fixes #776) by @enyst in #783
• Rename metadata field to litellm_extra_body and add custom config support by @li-boxuan in #837
• Fix error in 422 logs due to blank bearer token by @tofarr in #1022
• Fix abbreviation formatting in visualizer by @SmartManoj in #989
• Implement automatic tool registration on import by @malhotra5 in #862
• feat: Add automatic loading of user skills from home directory by @whywaita in #950
• Fix multi-arch manifest merge to use correct tag names by @xingyaoww in #1024
• Disable Copilot chat command center icon by default (parity with OpenHands PR #11589) by @enyst in #1032
• Fix Docker cache tag length exceeding 128 character limit by @xingyaoww in #1029
• Fix Laminar span stack warning in LocalConversation by @jpshackelford in #1039
• Include 03_browser_use_with_docker_sandboxed_server.py in test-examples workflow by @xingyaoww in #1021
• A couple of fixes for OpenAI o3 by @enyst in #1031
• refactor: standardize Observation base class by @simonrosenberg in #929
• Fix: Make reasoning_effort optional by @Chesars in #1004
• Simplify hello world example by @rbren in #1047
• Add reviewer trigger for PR review workflow by @simonrosenberg in #1037
• Fix RemoteConversation stats field mismatch and state updates by @xingyaoww in #1042
• Rename execute_bash to terminal & rename BashTool to TerminalTool by @SmartManoj in #1033
• Improve visualization API: allow passing ConversationVisualizer directly by @xingyaoww in #1025
• v1.0.0 release by @xingyaoww in #1056

New Contributors

• @YakshithK made their first contribution in #934
• @ApGa made their first contribution in #930
• @softpudding made their first contribution in #964
• @CLFutureX made their first contribution in #949
• @SmartManoj made their first contribution in #1013
• @whywaita made their first contribution in #950
• @Chesars made their first contribution in #1004

Full Changelog: 1.0.0a5...1.0.0

1.0.0a5

What's Changed

• Fix PR review workflow subdirectory paths by @simonrosenberg in #824
• Fix httpx.UnsupportedProtocol error by setting base_url on clients by @neubig in #801
• docs: set tags=["Server Details"] for / path by @xingyaoww in #829
• Enforce batch atomicity for condenser by @ryanhoangt in #775
• Replace All-Hands-AI references with OpenHands by @rbren in #830
• Fix: Make adding security_analyzer optional for get_default_agent by @jimwhite in #802
• Add test to reproduce bug #825: execute_command includes previous output by @neubig in #828
• Add FunctionCallValidationError handling & claude-haiku-4-5 to function calling supported models by @ryanhoangt in #832
• Set add_security_analyzer=False by default and simplify examples by @xingyaoww in #846
• Debug: glob tool fails when pattern contains path by @simonrosenberg in #718
• .github/workflows: Migrate workflows to Blacksmith runners by @blacksmith-sh[bot] in #831
• Rename "microagents" to "skills" throughout the repository by @neubig in #820
• Fix FastAPI path parameters to handle file paths with forward slashes by @amanape in #848
• Env template generator by @tofarr in #847
• Socket logging by @tofarr in #858
• Simplify examples for docs by @xingyaoww in #859
• fix: ReadTimeout when calling 01_convo_with_local_agent_server.py by @juanmichelini in #857
• Simplify remote agent server examples by @xingyaoww in #868
• Add conversation to tool calls by @simonrosenberg in #877
• Add workflow to check documented examples by @xingyaoww in #878
• chore(examples): env-driven endpoints, platform detection; update domains and API var names by @enyst in #872
• Fix DockerWorkspace execute_command timeout issue by @neubig in #871
• tests(remote): silence pytest collection warning by renaming helper class by @enyst in #880
• Fix agent status not transitioning to running status by @ryanhoangt in #884
• Dependency Injection for EventService by @tofarr in #885
• add asssign reviews workflow to examples by @simonrosenberg in #883
• feat(tests): Add comprehensive test suite for RemoteConversation implementation by @xingyaoww in #451
• Switch back to pyright + stricter typing to allow vscode to resolve imports by @xingyaoww in #869
• Fix pyright issues in agent status transition tests by @enyst in #895
• Fix MCP tool validation error by excluding 'kind' field from action data by @xingyaoww in #887
• Add regression test for bash command endpoint fix (#866) by @xingyaoww in #881
• Update sdk version to 1.0.0a4 by @xingyaoww in #901
• Fix PydanticSerializationUnexpectedValue warnings in log_completions by @xingyaoww in #904
• Fix #905: Handle None context_window for unmapped models by @xingyaoww in #906
• Add workflow to run examples script & fix failed integration tests by @ryanhoangt in #914
• AsyncRemoteWorkspace now has working directory by @tofarr in #888
• refactor: move agent_final_response to utility function with proper type checking by @xingyaoww in #919
• Feat encrypted secrets by @tofarr in #893
• chore: remove Dependabot config (uv workspace unsupported) by @enyst in #897
• Bump the version-all group with 3 updates by @dependabot[bot] in #913
• Add docs workflow to repo.md and fix pyright type checking by @xingyaoww in #924
• feat: Implement blocking delegation by @simonrosenberg in #908
• refactor: Update ToolExecutor interface to use LocalConversation instead of BaseConversation by @xingyaoww in #925
• Fix #916: Suppress Pydantic serializer warnings from litellm by @xingyaoww in #917
• Update README to refer to documentation site by @xingyaoww in #928
• Feat git operations by @tofarr in #863
• refactor: Clean up _configure_bash_tools_env_provider anti-pattern by @xingyaoww in #922
• Update integration test to use OpenHands proxy instead of litellm proxy by @simonrosenberg in #944
• feat: Post integration test results to tracker issue for daily runs by @simonrosenberg in #940
• Update litellm proxy path in workflows by @simonrosenberg in #946
• bump browser-use version, infer screenshot mime type dynamically by @dinmukhamedm in #935
• change base url in integration tests by @simonrosenberg in #947
• feat: support absolute paths for system_prompt_filename by @xingyaoww in #902
• Fix integration tests to work with fork PRs by @xingyaoww in #932
• refactor: fix naming inconsistency by @simonrosenberg in #958
• release version 1.0.0a5 by @xingyaoww in #959

New Contributors

• @jimwhite made their first contribution in #802
• @blacksmith-sh[bot] made their first contribution in #831
• @amanape made their first contribution in #848
• @dinmukhamedm made their first contribution in #935

Full Changelog: 1.0.0a3...1.0.0a5

1.0.0a3

What's Changed

• Add example for automated TODO management with GitHub Actions by @simonrosenberg in #758
• Rename LLM service identifiers to usage_id by @enyst in #799
• Bump the version-all group with 4 updates by @dependabot[bot] in #810
• Fix assign-reviews workflow subdirectory paths by @neubig in #807
• Fix websocket deprecation warning by using wsproto by @xingyaoww in #508
• Fix: Make Check OpenAPI Schema workflow work with fork PRs by @xingyaoww in #814
• Setting VSCode token to match SESSION_API_KEY. by @tofarr in #793
• Fix deprecated @model_validator usage in LLMSummarizingCondenser by @xingyaoww in #817
• Fix: Skip Docker build and push for fork PRs by @xingyaoww in #821
• 1.0.0a3 pre-release by @xingyaoww in #822

Full Changelog: 1.0.0a2...1.0.0a3

1.0.0a2

What's Changed

• Sync readme to changes by @enyst in #734
• use uv-managed python for "source" docker by @xingyaoww in #737
• Render user rejections in visualizer by @enyst in #746
• fix: make ThinkAction skip user confirmation like FinishAction by @malhotra5 in #742
• Add GitHub Actions workflow for scheduled maintenance tasks by @neubig in #672
• Enable unused variable detection in pre-commit hooks by @xingyaoww in #726
• Add optional conversation_id parameter to StartConversationRequest by @malhotra5 in #741
• Add AsyncRemoteWorkspace and sync with APIRemoteWorkspace by @tofarr in #735
• Update all SDK examples to read LLM_MODEL and LLM_BASE_URL from environment by @neubig in #740
• Add PR review workflow for fine-grained review comments by @simonrosenberg in #751
• Add "assign reviewers" workflow by @neubig in #756
• Update assign reviewers script based on experience by @neubig in #760
• Fix for URL endpoint location by @tofarr in #503
• Bump the version-all group with 2 updates by @dependabot[bot] in #535
• refactor: remove duplicated example and rename folder by @xingyaoww in #765
• Extend context truncation cases: broaden vLLM error detection by @enyst in #759
• Add support for claude-4-5-haiku by @ryanhoangt in #768
• Fix PyPI Wheel by @xingyaoww in #752
• Print conversation ID when an exception stops the run (fixes #777) by @enyst in #779
• Feat: Add Configurable Plan Template to Planning Agent by @simonrosenberg in #778
• refactor(llm): extract kwargs normalization into pure options selectors by @enyst in #643
• Add basedpyright to pre-commit and fix all type errors by @xingyaoww in #593
• fix: deploy changes to docs repo by @xingyaoww in #786
• update deploy-to-docs workflow by @xingyaoww in #788

Full Changelog: 1.0.0a1...1.0.0a2

1.0.0a1

What's Changed

• Setup Makefile, Pre-commit, and initial Tool and Schema definition by @xingyaoww in #1
• Port over file editor tool by @xingyaoww in #2
• Fix tool schema and add relevant tests; Add bash implmentations by @xingyaoww in #4
• Update README and add repo.md by @xingyaoww in #5
• ci: add coverage report and summary to tests workflow (closes #3) by @xingyaoww in #6
• Port over BashSession and tests by @xingyaoww in #7
• Openhands/move coverage badge to docs assets by @xingyaoww in #10
• Update ci badge on pr only by @xingyaoww in #11
• Add minimal config & port over the LLM class by @xingyaoww in #12
• Only trigger tests on PR by @xingyaoww in #15
• Port over minimal set of agent, config, context, microagent; adjust existing tool definition by @xingyaoww in #18
• Improve tool call visualization by @xingyaoww in #20
• run test on main by @xingyaoww in #21
• logger: set LiteLLM loggers to WARNING by default; enable DEBUG with DEBUG_LLM confirmation by @xingyaoww in #23
• Remove json repair by @xingyaoww in #24
• Add ruff configuration for import sorting by @xingyaoww in #17
• Simplify pre-commit CI by @xingyaoww in #26
• rename package openhands-ai to openhands-sdk by @xingyaoww in #27
• Rename openhands-ai as openhands-sdk by @xingyaoww in #28
• Rename unnecessary test_ruff_config.py by @xingyaoww in #29
• Restructure agent-sdk repository into UV workspace with simplified structure by @xingyaoww in #33
• add openhands setup.sh by @xingyaoww in #31
• ci: make coverage PR comment sticky to avoid duplicates (fixes #34) by @xingyaoww in #35
• refactor repo by creating context/; move looping logic into conversation.py by @xingyaoww in #36
• Update repo.md for tests by @xingyaoww in #37
• support multiple callback fn; improve examples of how to add callback from script by @xingyaoww in #38
• refactor FinishTool to builtin tools for agents by @xingyaoww in #39
• follow torvards-mode suggestion: integrate lock with state, and modify state in-place by @xingyaoww in #40
• Make unit test jobs conditional based on folder changes by @xingyaoww in #46
• refactor the system to use Event Types by @xingyaoww in #44
• Conversation: make state.events.append a default callback; remove manual appends in CodeActAgent by @xingyaoww in #57
• Remove prompt manager and instead using simpler jinja util function by @xingyaoww in #63
• Port over microagent changes in OpenHands/OpenHands#10528 by @xingyaoww in #65
• Add ruff format to pre-commit hooks for consistent code formatting by @xingyaoww in #67
• Fix: Include Jinja2 template files in package distribution by @malhotra5 in #69
• Reduce line-length from 500 to 88 and fix pre-commit errors by @xingyaoww in #68
• Add remaining LLM fixes and LLM registry by @ryanhoangt in #70
• Port LLM tests from OpenHands to agent-sdk by @enyst in #48
• Standardize project test structure by @xingyaoww in #75
• Standardize on absolute imports (PEP 8) and ensure repo package precedence by @xingyaoww in #74
• Rename CodeActAgent to Agent and update folder structure by @xingyaoww in #78
• Rename openhands.core to openhands.sdk by @xingyaoww in #73
• Core context condensation implementation by @csmith49 in #61
• test: Add comprehensive tests for events_to_messages conversions by @xingyaoww in #81
• Support microagents and unify it within AgentContext by @xingyaoww in #71
• feat: Add BashTool and FileEditorTool subclasses for simplified tool initialization by @xingyaoww in #88
• Fix coverage upload issue in CI workflow by @xingyaoww in #90
• Optimize pre-commit speed by setting always_run to false by @xingyaoww in #96
• fix typo in prompt suffix by @xingyaoww in #97
• Configure Dependabot for uv and GitHub Actions by @xingyaoww in #98
• fix: Minor test cleanup by @csmith49 in #99
• Bump astral-sh/setup-uv from 3 to 6 by @dependabot[bot] in #100
• Add open-pull-requests-limit and groups to dependabot by @xingyaoww in #105
• Bump the version-all group with 4 updates by @dependabot[bot] in #108
• Bump actions/checkout from 4 to 5 by @dependabot[bot] in #104
• Port config tests from OpenHands to agent-sdk by @enyst in #49
• fix: Idempotent Message serialization by @csmith49 in #106
• Bump actions/download-artifact from 4 to 5 by @dependabot[bot] in #101
• Implement confirmation mode for OpenHands agent SDK by @malhotra5 in #80
• feat(agent): simplify security guidelines in system prompt by @enyst in #115
• Update system prompt SECURITY section to match OpenHands PR #10822 by @xingyaoww in #114
• feat: improve visualizer by @xingyaoww in #119
• Migrate CLI subprocess shell and refactor BashSession architecture by @xingyaoww in #109
• Refactor LLMConfig into LLM pydantic class & large refactor of LLM class logic by @xingyaoww in #91
• Clean up all examples; add interactive bash example; improve visualizer for action by @xingyaoww in #122
• Update README.md and fix an AgentContext bug by @xingyaoww in #127
• Port runtime bash tests by @ryanhoangt in #84
• Migrate MCP clients by @ryanhoangt in #94
• feat(logger): improve logger using RichHandler by @xingyaoww in #132
• Fix confusing LLMRegistry API after LLM/LLMConfig merge by @xingyaoww in #129
• Add pause/resume functionality to agent-sdk conversation system by @malhotra5 in #118
• Add truncation to ExecuteBashObservation and TextContent with warning checks by @xingyaoww in #124
• docs: Add comprehensive tools system documentation by @xingyaoww in #136
• Move tests to a new directory by @xingyaoww in #140
• Fix pause event type by @malhotra5 in #144
• chore(docs): create docs/ folder and move existing documentation by @enyst in #117
• fix: resolve empty API keys to None and add Bedrock model support by @enyst in #141
• Fix prompt caching and improve input/output metrics visualization by @xingyaoww in #143
• Stabilize tests: telemetry logging warning and Bash PS2 prompt capture by @enyst in #145
• Fix issue #10729: Add xai/grok-code-fast-1 to MODELS_WITHOUT_STOP_WORDS by @enyst in #147
• Update microagent prompt to be more collaborative and less abrasive by @xingyaoww in #149
• feat: Lazy discriminated unions for event/action/observation serialization by @csmith49 in https://github....