NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
PyPI, #2885 most downloaded on PyPI
The programming language for agentic software.
Latest stable
3.2.0
Last release 8 days ago
12 Nov 2026
Ships on a steady schedule
a new release about every 9 days
Nearly every release is documented
notes for 59 of the last 60 stable releases
Nothing withdrawn
no release was ever pulled
2 years old
278 releases · first in 2024
Drag across the ruler to compare, or tap a release to open it.
Preserve Registered Model Params on Reconstruction: Reuse the live registry model instance when rebuilding DB-stored agents/teams so connection params
azure_endpoint/base_url and credentials
survive instead of being dropped to None. (#8476)Full Changelog: https://github.com/agno-agi/agno/compare/v2.6.17...v2.6.18
Resilient DB Component Loading: Isolate each agent/team load so one bad component is skipped instead of dropping all, fix model provider round-trip on
TuningEngines, and dedupe the catalog by model class. (#8461)(type, name, function set) so they collapse instead of accumulating duplicate registry entries. (#8450)xoxp-) for search_messages, which bot tokens can't call. (#8411)required rows aren't orphaned. (#8386)Full Changelog: https://github.com/agno-agi/agno/compare/v2.6.16...v2.6.17
Your coding agent can read these notes before it upgrades. Set up the MCP server →
Support parallel-web >= 1.0 GA API in ParallelBackend: migrate web_search/web_extract to the top-level client and pin the floor to >=1.0
Full Changelog: https://github.com/agno-agi/agno/compare/v2.6.15...v2.6.16
Custom, Scoped, Identity-Aware MCP Tools: The AgentOS MCP server (served at /mcp) is now a real extension point, configured through a single MCPServer
MCPServerConfig object. You can register custom tools (plain callables or Agno @tool/Functions), scope the built-ins (enable_builtin_tools=False, or filter with include_tags/exclude_tags), inject the authenticated caller identity into a tool (declare user_id and AgentOS supplies the JWT subject while hiding it from the client schema), gate calls with a one-line authorize function, and opt into built-in DNS-rebinding protection via allowed_hosts/allowed_origins. All with data, no custom middleware classes. Fully backward compatible: without mcp_config, all built-in tools register exactly as before.Full Changelog: https://github.com/agno-agi/agno/compare/v2.6.14...v2.6.15
Learnings CRUD on AgentOS: Added create, read, update, and delete endpoints for learnings on AgentOS.
json_object providers.Full Changelog: https://github.com/agno-agi/agno/compare/v2.6.13...v2.6.14
Sub-Agent Event Streaming: Sub-agent events from the context provider update tool now stream through to the parent run.
json_schema now handles Optional dataclass fields that have no declared type.MCPTools sessions are now closed in the call task.upsert=False inserts of the same document no longer collapse.DaytonaTools.upsert=False inserts of the same document don't collapse by @sannya-singal in https://github.com/agno-agi/agno/pull/8310Full Changelog: https://github.com/agno-agi/agno/compare/v2.6.12...v2.6.13
HTML File Generation: Added HTML file generation support with example app.
drop_ratio_search and remove drop_ratio_buildClient.results() usage.IndexError when runs list is empty in from_dict.enable_agentic_state tool receives correct schema for dict params.ag-ui-protocol>=0.1.14 to prevent reasoning role validation error.Client.results() in ArxivReader by @sannya-singal in https://github.com/agno-agi/agno/pull/8261Full Changelog: https://github.com/agno-agi/agno/compare/v2.6.11...v2.6.12
Parallel Web Task API & Monitor API Tools: Added tools for Task API and Monitor API integration for parallel web.
Manifest for per-entity AgentOS UI metadata configuration.v25.0.Full Changelog: https://github.com/agno-agi/agno/compare/v2.6.10...v2.6.11
Replace deprecated asyncio.get_event_loop() with get_running_loop() in embedders.
YouTools for You.com Search API integration.files field on RunCompleted event.google-interactions provider to model string parser.team/agent/run_context params.Attachment.read() instead of blocking requests.get in Discord client.time.sleep with await asyncio.sleep in _async_create_collection_and_scope.asyncio.get_event_loop() with get_running_loop() in embedders.CSVReader.async_read uses newline join to match sync read behavior.OpenAITools.transcribe_audio._acontinue_run_background_stream to prevent AttributeError on continue-run SSE.parser_model is set (Agent and Team); added parser_model guard to continue_run_dispatch and acontinue_run_dispatch.contents_db; removed deprecated Tantivy FTS, fixed Redis hset mypy error.fal.py.memory_topics signature mismatch across DB backends.ValueError in delete_session when called with async db.team/agent/run_context params by @sannya-singal in https://github.com/agno-agi/agno/pull/8066time.sleep with await asyncio.sleep in _async_create_collection_and_scope by @kratos0718 in https://github.com/agno-agi/agno/pull/8158OpenAITools.transcribe_audio (resource leak) by @kratos0718 in https://github.com/agno-agi/agno/pull/8161Full Changelog: https://github.com/agno-agi/agno/compare/v2.6.9...v2.6.10
decision_log - drop deprecated `datetime.utcnow()`: Replaced with datetime.now(timezone.utc) to clear the Python 3.12 deprecation warning. (Closes #80…
resolved_by, resolved_at, etc.) via run_response.metadata["approval"]. Previously only status and resolution_data were exposed. Lives in metadata so it works uniformly across RunOutput / TeamRunOutput / future WorkflowRunOutput. (#7366, #8032)PgVector(prefix_match=True) was a silent no-op - it appended * then routed through websearch_to_tsquery, which ignores wildcards. Now routes through to_tsquery('tok:*') with proper tokenization, so partial queries like "ani" FTS-match "animal" as documented. New cookbook cookbook/07_knowledge/04_advanced/06_prefix_search.py shows the help-center typeahead use case. (#8048, #8051, #8052)to_tsquery(language, '') fallback with a literal ''::tsquery cast so behavior no longer depends on the parser tolerating empty input. (#8053)0.0 values due to bare truthiness checks. Switched to is not None so deterministic output works as intended. (#8009, fixes #8004)DEFAULT_READ_INSTRUCTIONS / DEFAULT_WRITE_INSTRUCTIONS from 157 lines to 27 by removing content already supplied by the toolkit-side instructions. Provider now owns role + safety; toolkit owns the technical reference. (#7982)datetime.utcnow(): Replaced with datetime.now(timezone.utc) to clear the Python 3.12 deprecation warning. (Closes #8030)>=1.6.7 to pick up the upstream language auto-detection fix; removed the temporary test workarounds from #7904. (#8012)FunctionCallSteps describe tools the autonomous loop runs inside the server-managed sandbox. Previously we surfaced them as local tool_calls, leading to Function <name> not found and follow-up 400 invalid_request errors. Gated the FunctionCallStep branch on self.agent is None for both streaming and non-streaming. Model path with user-declared tools is unchanged. (#8045)temperature=0 silently dropped: See Improvements. Every prior release silently fell back to API defaults (~1.0) when callers explicitly set 0. (#8009, fixes #8004)>=1.6.7 by @sannya-singal in https://github.com/agno-agi/agno/pull/8012Full Changelog: https://github.com/agno-agi/agno/compare/v2.6.8...v2.6.9
Introduced PathSecurityError (FileGenerationSecurityError kept as deprecation alias). Bumped requires-python to >=3.9,<4.
AntigravityAgent (a BaseExternalAgent served through AgentOS with native sessions/streaming/UI) and AntigravityTools (a Toolkit that lets any Agno agent delegate a sub-task to a managed Antigravity sandbox).GeminiInteractions support for Google's managed agents — Deep Research (autonomous research with citations, background streaming with reconnect/last_event_id resume) and Antigravity (general-purpose agent in a managed Linux sandbox). New agent / agent_config / environment fields, per-agent forcing of background and store, and support for mcp_servers + file_search_store_names on the agent path.cookbook/data_labeling/ with 18 self-contained workflows covering text, image, audio, video, document, and composed (LLM-as-judge, quality review) labeling primitives.tool_choice="required", user_input echo, and clean termination via stop_after_tool_call=True.agno.utils.path_safety module with safe_join and safe_join_subpath. Hardened FileGenerationTools, SlackTools, Toolkit._check_path, agno.skills.utils.is_safe_path, and FileTools.check_escape against path traversal, symlink escape, control-char injection, Windows MagicDot, and Unicode normalization attacks. Introduced PathSecurityError (FileGenerationSecurityError kept as deprecation alias). Bumped requires-python to >=3.9,<4.ParallelMCPBackend now sends User-Agent: agno/<version> on every request so Parallel can attribute traffic to agno.available_models field from EvalsDomainConfig; the top-level AgentOSConfig.available_models is now the only supported source for the Evals UI dropdown.gemini-3-flash-preview to gemini-3.5-flash across the Gemini Interactions cookbooks.stream_steps → stream_events) that caused teams running via AG-UI to silently drop all intermediate streaming events (lifecycle, reasoning, tool calls, member delegations) for ~7 months.aiohttp>=3.7.3,<4 to the slack optional extra — AsyncWebClient imports aiohttp at module load, so pip install agno[slack] was broken for all Slack operations.incompleteSearch flag from GoogleDriveTools.search_files() so corpora="allDrives" callers can detect when Drive did not search every drive.generation_config passthrough field for advanced controls (top_k, presence_penalty, etc.) and fixed history handling — when previous_interaction_id is set, only the new turn is sent rather than replaying the full history. Now actually leverages server-side statefulness.server_tool_use and code-execution content blocks in message history so multi-turn server-tool flows no longer break across turns.redacted_thinking type literal (the SDK rejects redacted_reasoning_content), accepted both spellings on the streaming start event, added a dedup guard on round-tripped server tool blocks, made coercion failures visible via log_warning, and coerced non-block tool-result items to text.str(error) is empty or weak, preserving error context across invoke / invoke_stream / ainvoke / ainvoke_stream.n1n provider mapping in get_model() so Agent(model="n1n:...") model-string paths now resolve correctly.Full Changelog: https://github.com/agno-agi/agno/compare/v2.6.7...v2.6.8
Gemini Interactions: Added a new GeminiInteractions model class that makes use of Googles new stateful interactions API
GeminiInteractions model class that makes use of Googles new stateful interactions APIallowed_hosts parameter to URL-fetching readers to restrict outbound fetches.async_insert.session_id / agent_id / team_id from being overwritten by a child agent's spans when both share a trace_id. Most visible when a Team uses a post-hook (e.g. @hook(run_in_background=True).acleanup_run if running in an async context.async_insert by @sannya-singal in https://github.com/agno-agi/agno/pull/7893allowed_hosts SSRF guard to knowledge readers by @sannya-singal in https://github.com/agno-agi/agno/pull/7892Full Changelog: https://github.com/agno-agi/agno/compare/v2.6.6...v2.6.7
Slack Interface: Added support for HITL multi-row approvals with all pause types.
NotionDatabaseBackend to wiki context provider.aget_last_run_output returns None when agent.id is auto-generated during arun()./continue Endpoint: Fixed to forward dependencies and metadata to /continue endpoints via get_request_kwargs.aget_last_run_output returns None when agent.id is auto-generated during arun() by @kausmeows in https://github.com/agno-agi/agno/pull/7840tree-sitter-language-pack<1.8.0 to unblock chonkie code chunker by @sannya-singal in https://github.com/agno-agi/agno/pull/7869Full Changelog: https://github.com/agno-agi/agno/compare/v2.6.5...v2.6.6
Gemini Multimodal File Search: Added multimodal support in the Gemini File Search API (for google-genai≥1.75.0). See cookbook.
google-genai≥1.75.0). See cookbook.GmailContextProvider and CalendarContextProvider following the same pattern as existing providers (GDrive, Slack, Database). Also updates GDriveContextProvider to support OAuth in addition to service account auth.on_error handling to the Condition workflow step, allowing users to control how errors are handled when sub-steps within a condition fail.allowed_hosts param so agents only fetch from hosts you trust.enable_media_tools flag (default: False) to control file download/upload.
download_file added to read toolsupload_file added to write toolsuser_id on AgentOS MCP tool handlers to the authenticated JWT subject to prevent cross-tenant reads and writes.@tool(instructions=...) to take effect when used inside a Toolkit.describe_table.GoogleDriveTools.update_memory and delete_memory tools.user_id to JWT subject by @sannya-singal in https://github.com/agno-agi/agno/pull/7811Full Changelog: https://github.com/agno-agi/agno/compare/v2.6.4...v2.6.5
WikiContextProvider (#7720): New context provider with filesystem and git backends, web ingestion, and read/write flags.
WikiContextProvider (#7720): New context provider with filesystem and git backends, web ingestion, and read/write flags.Full Changelog: https://github.com/agno-agi/agno/compare/v2.6.3...v2.6.4
WorkspaceContextProvider (#7709): Project-aware context provider for repository roots, backed by the read-only Workspace toolkit instead of generic Fi
groups:read scope is missing.Full Changelog: https://github.com/agno-agi/agno/compare/v2.6.2...v2.6.3
Models: Updated default id for multiple model providers to newer, actively supported models to prevent upcoming deprecations, improve performance cons…
read/list/search/write/edit/move/delete/shell access to a root directory tree, with destructive operations gated by Agno's built-in human-in-the-loop confirmation by default.id for multiple model providers to newer, actively supported models to prevent upcoming deprecations, improve performance consistency, and significantly reduce operational costs.deep_copy for Step, Router, and Loop.Full Changelog: https://github.com/agno-agi/agno/compare/v2.6.1...v2.6.2
A2A: Pinned a2a-sdk>=0.3.0,<1.0 in the release workflow. a2a-sdk v1.0 has breaking changes that need to be adopted in a follow-up.
system_prompt_blocks: List[SystemPromptBlock] on Claude — each block has text, cache, and optional ttl ("5m" or "1h") that overrides the model-level extended_cache_time flag per block.cache_tools: bool on Claude (Anthropic, AWS Bedrock, VertexAI). Adds cache_control to the last tool so the tool prefix is cached.Model._format_tools (sort by name), so request prefixes stay stable across runs and prompt caches actually hit. Applies across Anthropic, OpenAI, Gemini, and Bedrock.ParallelMCPBackend as a new web backend for WebContextProvider (#7667). Talks to Parallel's public MCP server at search.parallel.ai/mcp, exposing web_search + web_fetch (compressed markdown output). Keyless by default; Bearer-auth via PARALLEL_API_KEY for higher rate limits; optional OAuth endpoint via use_oauth=True. Defaults to a 30s timeout (up from MCPTools' 10s) because web_fetch frequently runs longer on large pages."openai:" model string prefix to OpenAIResponses (previously OpenAIChat). Added "openai-chat:" as a fallback for users who still need OpenAIChat. Agent(model="openai:gpt-5.4") now resolves to OpenAIResponses(id="gpt-5.4").a2a-sdk>=0.3.0,<1.0 in the release workflow. a2a-sdk v1.0 has breaking changes that need to be adopted in a follow-up.output_schema example to demonstrate output_schema with a Pydantic model (it was incorrectly using output_model with web search).cookbook/05_agent_os/README.md that referenced a non-existent agno_agent.py (corrected to agno_assist.py) (#7663).Full Changelog: https://github.com/agno-agi/agno/compare/v2.6.0...v2.6.1
Team HITL: Added API layer for Team human in the loop and support in AgentOS chat page.
AgentOS chat page.AgentOS chat page.ClaudeAgentSDK, Langgraph and DSPy using a single AgentProtocol as the backbone which works with AgentOS runtime.AgentFactory, TeamFactory and WorkflowFactory allowing for more multi-tenant usecases.agno.context — a first-party API for plugging an external source (filesystem, web, SQL database, Slack, Google Drive, MCP server) into anagent as a natural-language tool.Agent.deep_copy.GET /workflows/{id} which omitted nested workflow steps and Condition.else_steps (along with the agents inside them) from the responseparent_step_id through nested workflow enrichment layers in case of workflow as a workflow step.agno.utils.models.claude.content_hash values.exclude_patterns to FileTools./sessions endpoint now returns all session types (agent, team, and workflow) by default, giving you a complete view of your sessions in a single call. To filter for a specific type, pass?type=agent ?type=team, or ?type=workflow in your request.Option to disable Claude file citations
GitHubConfig repo to be specified per request (#7496)header_provider headers during MCP initialization (#7507)nested_depth to agent/team events (#7491)CancelledError explicitly in all router streaming generators (#7379)user_input_schema (#7485)extra_messages in memory pipeline gate check (#7470)Full Changelog: https://github.com/agno-agi/agno/compare/v2.5.16...v2.5.17
LLMsTxtTools and LLMsTxtReader: Adds support for the llms.txt standard - a standardized way for websites to provide LLM-friendly documentation indexes
https://docs.agno.com/llms.txt).knowledge_table from contents_db instead of agent.db.response.reasoning_summary_text.delta events to stream reasoning content as it arrives.TeamSession.from_dict() to not mutate input mapping.Full Changelog: https://github.com/agno-agi/agno/compare/v2.5.15...v2.5.16
Skills: Added skills support to Team. See docs.
HumanReview config class - pass human_review=HumanReview(...) on Step, Loop, and Router instead of flat params. Fully backward compatible. See docsAGNO_LOG_TRACEBACKS env var (opt-in) to control traceback visibility in log_error and log_warning — off by default (clean logs), full tracebacks when set to true.last_n_runs and conversation_limit parameters to
SessionSummaryManager so users can control how much conversation history is included when generating session summaries.400 errors under concurrent usage by letting the OpenAI SDK manage its own HTTP client.audio_total_tokens not being computed for OpenAI, Perplexity, and LiteLLM.Condition step type.output_schema content before str concatenation.TeamSession.get_messages(..) returning duplicate messages.get_pull_requests when fewer PRs than requested limitCondition step type by @korntewin in https://github.com/agno-agi/agno/pull/7198TeamSession.get_messages return duplicated messages by @kausmeows in https://github.com/agno-agi/agno/pull/7372Full Changelog: https://github.com/agno-agi/agno/compare/v2.5.14...v2.5.15
Fallback Models: Added support for Fallback Models on Agents and Teams. See docs `python agent = Agent( model=OpenAIChat(id="gpt-4o", base_url="http:/
agent = Agent(
model=OpenAIChat(id="gpt-4o", base_url="http://localhost:1/v1", retries=0),
fallback_models=[Claude(id="claude-sonnet-4-20250514")],
)
AzureBlobConfig.SlackTools.inject_trailing_user_message and `trailing_user_message_content for models that reject assistant prefill by @harshsinha03 in https://github.com/agno-agi/agno/pull/7304Full Changelog: https://github.com/agno-agi/agno/compare/v2.5.13...v2.5.14
ReliabilityEval: Add subset matching, argument validation, and missing tool call tracking with multi-round tool call collection fixes.
messages parameter in _prepare_request_kwargs for Claude sub classes/info api to return agent, team, workflow count by @kausmeows in https://github.com/agno-agi/agno/pull/7190Full Changelog: https://github.com/agno-agi/agno/compare/v2.5.12...v2.5.13
fix: preserve server tool blocks in Claude conversation history by @Mustafa-Esoofally in https://github.com/agno-agi/agno/pull/6879
Full Changelog: https://github.com/agno-agi/agno/compare/v2.5.11...v2.5.12
Google Sides Toolkit: Added new slides toolkits for creating / editing and working with Google slides
GoogleAuth toolkit and shared auth decorator for Google toolkits.AgenticChunkingPerplexitySearch toolkit.run_status column valid for ≥v2.5.6search() in order to avoid hybrid search returning the same document from both vector and FTS branches.0.1.x to 0.2.0.output_config request parameter for Claude wrappers, so users can pass values like output_config={"effort": "high"} together with adaptive thinking.$defs in Claude tool schema formattingfile_ids and container info in streaming path for Claude skillsparse_tool_calls.AWS_BEDROCK_API_KEY auth in Claude.Full Changelog: https://github.com/agno-agi/agno/compare/v2.5.10...v2.5.11
Parallel Search Tool: Add Parallel AI Search support for Vertex AI with native ToolParallelAiSearch integration
/new to start fresh conversations.enable_encryption parameter for encrypting phone numbers/new to start fresh conversations.version query param to GET /workflows/{id}to fetch specific workflow versions.metadata, dependencies, add_dependencies_to_context, add_session_state_to_context similar to agent/team.mistralai v2 support while maintaining backward compatibility with v1.MCPTools: Fixed race condition where parallel MCP tool calls with header_provider would each create duplicate sessions, causing the agent to get stuck.parse_tool_callsFull Changelog: https://github.com/agno-agi/agno/compare/v2.5.9...v2.5.10
Followups: Added built-in followup suggestions for Agent and Team. See cookbook.
datetime_format parameter to Agent and Team, allowing custom strftime formatting for datetime context (e.g., ISO-8601, date-only, localized).run_context.messagesGoogleCalendarTools with new tools, service account auth, and cookbooks.add_history_to_context during multi-round conversations while runningbase_url from .com to .cnLearningMode.PROPOSE for multi-turn confirmation.Form/File parameters across agent, team, and workflow API endpoints for OpenAPI/Swagger visibilityFull Changelog: https://github.com/agno-agi/agno/compare/v2.5.8...v2.5.9
Human-Readable IDs: Agents and teams now use human-readable identifiers (e.g., brave-falcon-7x3k) instead of UUIDs, making debugging and monitoring mo
brave-falcon-7x3k) instead of UUIDs, making debugging and monitoring more intuitive.serve() now supports AGENT_OS_HOST and AGENT_OS_PORT environment variables as fallbacks, simplifying container deployments.GitlabTools with read-focused GitLab integrations, async support, and cleaner tool configuration.GmailTools with new tools and service account auth.store_media setting and is rather scrubbed before storing in DBexternal_execution and regular tools when using OpenAI Responses API.json_serializer to MySQL engine creation for proper JSON handling.session_type filter from get_session to prevent session data loss when session types change.Loop iterations always received the original input instead of the previous iteration's output. Added a flag forward_iteration_output to opt-in to this behaviour.Full Changelog: https://github.com/agno-agi/agno/compare/v2.5.7...v2.5.8
OpenAI-Like Embedder: Added OpenAILikeEmbedder for providers with OpenAI compatible embedding endpoints
OpenAILikeEmbedder for providers with OpenAI compatible embedding endpointssearch_past_sessions + read_past_session pattern. Session previews show per-run user/assistant pairs instead of a single message, with configurable limits on sessions searched (num_past_sessions_to_search) and preview depth (num_past_session_runs_in_search). read_past_session accepts an optional num_runs parameter so the model can avoid pulling in full conversations from long sessions.user_id from being propagatedFull Changelog: https://github.com/agno-agi/agno/compare/v2.5.6...v2.5.7
GitHub App Authentication for Knowledge Sources: Adds support for authenticating with GitHub as a GitHub App (app_id, installation_id, private_key) in
app_id, installation_id, private_key) in addition to personal access tokens, with thread-safe token caching and both sync/async variants.image/heic and image/heif support to file upload endpoints.agno.tools.google sub-package (from agno.tools.google import GmailTools). Old import paths still work with a backwards compatibility.TeamMode.tasks streaming to provide structured task data for frontend rendering in events like TeamTaskCreated and TeamTaskUpdatedTaskStateUpdatedEvent: Added tasks: List[TaskData] field containing full task details (id, title, description, status, assignee, dependencies, result) and completion_summary fieldfile_search silently returned empty results due to eventual consistency in OpenAI's vector_stores.files.list
API.input_file support to OpenAIResponses_rerank() call instead of being reused.Literal types in JSON schema generationFull Changelog: https://github.com/agno-agi/agno/compare/v2.5.5...v2.5.6
Responses now stream in real-time with live progress cards for tool calls, reasoning, and workflow steps.
token and signing_secret, enabling multiple bots on the same server.ModelsLabTools to support image generation via ModelsLab's text-to-image API, completing the full ModelsLab media suite.toolResult blocks into single user message._convert_image_artifacts_to_imagesFull Changelog: https://github.com/agno-agi/agno/compare/v2.5.4...v2.5.5
PgVector: added similarity_threshold parameter for filtering search results by minimum similarity score.
similarity_threshold parameter for filtering search results by minimum similarity score.TeamMode.tasks, enabling real-time event emission during autonomous task execution.DuckDuckGoTools@tool decorator: Fixed @tool decorator to not fail for Agent/Team type annotationsFull Changelog: https://github.com/agno-agi/agno/compare/v2.5.3...v2.5.4
Remote Knowledge: Added endpoints to support listing of remote contents via API. Enables uploading of content in S3 buckets via AgentOS
sanitize_content parameter to BasePDFReader (enabled by default) that normalizes fragmented PDF text extraction — collapsing word-per-line artifacts while preserving paragraph breaks. Disable with sanitize_content=False for PDFs containing structured content like code or tables.sanitize_content parameter to BasePDFReader that normalizes extracted PDF text by collapsing excessive whitespace sequences into single spaces.knowledge_retriever ignored when search_knowledge=True enabled.input_tokens for Anthropic modelsreasoning_content duplication across reasoning steps.deep_copy fails for nested teams due to non-init dataclass fields.session_state changes in both agent and team modules.run_context.output_schema in Parallel execution.Full Changelog: https://github.com/agno-agi/agno/compare/v2.5.2...v2.5.3
License Change: Changed license from Mozilla Public License 2.0 (MPL 2.0) to Apache Software License.
Full Changelog: https://github.com/agno-agi/agno/compare/v2.5.1...v2.5.2
TavilyTools: Removed deprecated include_answer kwarg from search method.
id over name in get_member_id for more predictable member selection.include_answer kwarg from search method.header_provider is incompatible with transport type.Full Changelog: https://github.com/agno-agi/agno/compare/v2.5.0...v2.5.1
Team Execution Modes: Introduced TeamMode enum with four execution modes:
TeamMode enum with four execution modes:
coordinate (default): Supervisor pattern where the leader picks members, crafts tasks, and synthesizes responsesroute: Router pattern where the leader routes to a specialist and returns their response directlybroadcast: Broadcast pattern where the leader delegates the same task to all members simultaneouslytasks: Autonomous task-based execution where the leader decomposes goals into a shared task listRunRequirement system for paused runs with better support for:
@approval decorator for human approval workflows:
@approval(type="required"): Blocking approval that pauses the run until resolved via approvals API@approval(type="audit"): Non-blocking audit trail for compliance/loggingLearningMachine for persistent learning.RunContext. Supports caching via cache_callables and custom cache keys for session-scoped resolution.isolate_vector_search flag to Knowledge class. When enabled, documents are tagged with linked_to metadata during insert and searches filter by this tag, allowing multiple Knowledge instances to share the same vector database while maintaining isolated search results. Defaults to False for backwards compatibility with existing data.search() to stop reopening the table; use .to_list() instead of .to_pandas() in vector/hybrid/keyword search.sessions, component_configs and component_links table to use primary key instead of _unique_constraints.store_history_messages now defaults to False: History messages are no longer stored by default. This reduces unnecessary storage overhead, improves privacy by not persisting conversation history unless explicitly needed, and gives better control over storage costs. If you rely on stored history, set store_history_messages=True explicitly.str() for workflow run_input by @salimhamed in https://github.com/agno-agi/agno/pull/6318Full Changelog: https://github.com/agno-agi/agno/compare/v2.4.8...v2.5.0
LanceDB: Added backward compatibility for deprecated table_names()
table_names()azure_ad_token_providerupdate_session_state tool to correctly extract session_state from run_context after _session_state was removed from the Function class in Agno 2.4.0.SessionContextStore for Claude/Gemini compatibility.session_id to executor events for consistent streaming by @kausmeows in https://github.com/agno-agi/agno/pull/6272Full Changelog: https://github.com/agno-agi/agno/compare/v2.4.7...v2.4.8
LanceDB - Replaces deprecated LanceDB table_names() with list_tables()
AwsBedrockReranker class added supporting Cohere Rerank 3.5 and Amazon
Rerank 1.0requires_confirmation_toolstable_names() with list_tables()else_steps in workflow condition logic by @uzaxirr in https://github.com/agno-agi/agno/pull/6214Full Changelog: https://github.com/agno-agi/agno/compare/v2.4.6...v2.4.7
Learning Machine: Improved prompts and default support for user_memory when using learning machine
user_memory when using learning machinecreated_at and updated_at timestamps now correctly convert to UTCFull Changelog: https://github.com/agno-agi/agno/compare/v2.4.5...v2.4.6
SeltzTools: A new toolkit featuring support for Seltz Search
add_search_knowledge_instructions: Adds add_search_knowledge_instructions parameter to Agent and Team class to control whether knowledge search instructions are added to the system prompt.`Full Changelog: https://github.com/agno-agi/agno/compare/v2.4.4...v2.4.5
Unsplash image search tools: Added UnsplashTools toolkit for searching and retrieving high-quality, royalty-free images from Unsplash API.
UnsplashTools toolkit for searching and retrieving high-quality, royalty-free images from Unsplash API.external_execution_silent param for tool decorator: Added a new param to make external tool execution silent by not populating the run response content with placeholder stringsdependencies in Workflows: When using JWTMiddleware with dependencies_claims to inject user data from JWT tokens, the dependencies were correctly available to agent tools when calling the agent directly via /agents/{agent_id}/runs, but not when the same agent ran as part of a workflow via /workflows/{workflow_id}/runs.split_on_headings now respects chunk_size parameter.created_at timestamps are now preserved on updates and get_session correctly filters by session typesilent option to suppress verbose messages for HITL tools by @rodrigocoliveira in https://github.com/agno-agi/agno/pull/5842Full Changelog: https://github.com/agno-agi/agno/compare/v2.4.3...v2.4.4
Excel Reader: Added new reader class ExcelReader for better .xls and .xlsx support
ExcelReader for better .xls and .xlsx supportCustomEvent :
acontinue_run for HITLFull Changelog: https://github.com/agno-agi/agno/compare/v2.4.2...v2.4.3
fix: Remove deprecated Pydantic Config and use model_config by @benmezger in https://github.com/agno-agi/agno/pull/6123
output_fields in document lookup queries to fix Milvus Lite errorNone values were incorrectly cached, preventing table creation on subsequent calls.model_config by @benmezger in https://github.com/agno-agi/agno/pull/6123Full Changelog: https://github.com/agno-agi/agno/compare/v2.4.1...v2.4.2
N1N Model Provider: Added n1n.ai as an OpenAI-compatible provider.
(.xlsx/.xls) by routing spreadsheets through the existing CSV reader, parsing workbooks per sheet into separate documents with sheet metadata.collect_metrics_on_completion flag to only collect metrics from the final chunk.tool_call_id to trace events back to their originating tool calls.Team.adelegate_task_to_members() where loop variables were not properly captured in async function definitions, causing all concurrent member agent runs to use the last member's values instead of their respective values.adelegate_task_to_members by @d8rt8v in https://github.com/agno-agi/agno/pull/6067Full Changelog: https://github.com/agno-agi/agno/compare/v2.4.0...v2.4.1
Deprecated enable_user_memories : Introduced update_memory_on_run for the same behavior.
KnowledgeProtocol The main Knowledge implementation now satisfies this interface. Any other implementation may be used with Agent/Team. Note - Only the main Agno Knowledge implementation of this protocol currently supports AgentOS Knowledge management.ModelRequestStarted and ModelRequestCompletedCompressionStarted and CompressionCompletedMemoryUpdateCompleted to contain the memory content.api_base_url parameter support to TavilyTools for custom hosted usecases.AgentOS: Add db to the AgentOS class. If populated, it propagates to all
agents, teams and workflows without db. It is also used as tracing db.enable_user_memories : Introduced update_memory_on_run for the same behavior.TrailingSlashMiddleware that strips trailing slashes from request paths, all endpoints returns identical responses without 307 redirects.<instructions> tags by defaultadd_instruction_tags=True option to restore the previous behavior for users who prefer structured promptsencoding parameter for BytesIO inputs - always used hardcoded "utf-8" instead of self.encoding or "utf-8"."") were incorrectly treated as None due to Python's falsy evaluationreasoning_details field when using OpenRouter, fixing multi-turn conversation issues.async_search() to use async_get_embedding() instead of blocking get_embedding() calls, preventing event loop blocking during concurrent async operations.session_state, dependencies, user_id etc in tool functions and hooks where RunContext has replaced itstream_intermediate_steps → stream_eventsyield_run_response → yield_run_outputdelegate_task_to_all_members removed from Team classadd_content() method and its variants have been renamed to insert() insert_many Old method names are still supported, but all examples and docs will soon only reference the new methods.tracing_db : tracing_db has been deprecated in favor of db parameter on the AgentOS.openai-agents install instruction by @harshsinha03 in https://github.com/agno-agi/agno/pull/6076Full Changelog: https://github.com/agno-agi/agno/compare/v2.3.26...v2.4.0
Per-Request Isolation: Improved request-level isolation for agents/teams/workflows in shared FastAPI processes.
Full Changelog: https://github.com/agno-agi/agno/compare/v2.3.25...v2.3.26
LearningMachine: A unified learning system that enables agents to learn from every interaction. It coordinates multiple learning types, each with its
get_page_content() tool failing on large pages by parsing HTML to extract only visible text content (strips scripts, styles, comments)chunk_size ParameterFull Changelog: https://github.com/agno-agi/agno/compare/v2.3.24...v2.3.25
MongoDB connection handshake now includes Agno version metadata. This enables better connection identification and log analysis for users hosting Mong
proxy_config parameter, allowing further configuration of the Toolkit.PythonTools and MLXTranscribeTools to automatically disallow operating outside of the contextual base directory. Add the restrict_to_base_dir to opt out of this feature and allow operations outside of the base directory.split_on_headings parameter to finely control how chunks are separated.Knowledge . Improved filter handling for agentic search.get_session_name directly from Team instances.Full Changelog: https://github.com/agno-agi/agno/compare/v2.3.23...v2.3.24
Async Tool Support: Toolkit now supports async tool functions, which are automatically selected when the agent runs in an async context.
header_provider : Fix for check on MCPTools.Full Changelog: https://github.com/agno-agi/agno/compare/v2.3.22...v2.3.23
Agent Skills: Introducing the Skills class, enabling you to extend your Agents’ capabilities by providing specific skills, as defined by Anthropic’s A
Skills class, enabling you to extend your Agents’ capabilities by providing specific skills, as defined by Anthropic’s Agent Skill specification. See the docs.header_provider function to your MCPTools instances to generate dynamic headers when calling MCP tools. This is useful for headers that need to change based on context e.g. authorization tokens or user ids. See the docs.audience for checkingMCPTools class now defaults to StreamableHttp as transport when connecting to external MCP servers.a2a protocol when using remote agents. The also means Google ADK agents can be used with AgentOS as remote running agents. This feature is still in beta. See the docs.stop flag from primitive inner steps to workflowCompressionManager to work properly with any model using the OpenAIResponses API.Full Changelog: https://github.com/agno-agi/agno/compare/v2.3.21...v2.3.22
Agent as Judge on AgentOS: Our new [AgentAsJudge Evals](https://docs.agno.com/basics/evals/agent-as-judge/overview) are now completely supported on th
Full Changelog: https://github.com/agno-agi/agno/compare/v2.3.20...v2.3.21
Run Cancellation: Add async methods to our Run Cancellation setup, and allow users to set a custom one using set_cancellation_manager().
set_cancellation_manager().reasoning_content for LiteLLM model wrapper: Extract reasoning content from the models which support it via lite llmretries set to more than 1.updated_at values.updated_at from pre 2.0 versions by @kausmeows in https://github.com/agno-agi/agno/pull/5824Full Changelog: https://github.com/agno-agi/agno/compare/v2.3.19...v2.3.20
Vector DB imports: Improved import paths for vector database integrations
Full Changelog: https://github.com/agno-agi/agno/compare/v2.3.18...v2.3.19
Google VertexAI: Added support for a google oauth2 credentials file for direct VertexAI authentication
Full Changelog: https://github.com/agno-agi/agno/compare/v2.3.17...v2.3.18
ChromaDB Hybrid Search: Added support for hybrid search using dense vector similarity search (semantic) with full-text search (keyword/lexical) using
RemoteAgent, RemoteTeam and RemoteWorkflow classes enabling usage of Agents, Teams and Workflows running on a remote AgentOS. Docs coming soon.AgentOSClient class enabling usage of an AgentOS running remotely. Docs coming soon.Full Changelog: https://github.com/agno-agi/agno/compare/v2.3.16...v2.3.17
Readers: Fixes an issue with read and async_read function of some Reader classes
read and async_read function of some Reader classesFull Changelog: https://github.com/agno-agi/agno/compare/v2.3.15...v2.3.16
OpenRouter Cost: Now tracking cost on metrics for runs on OpenRouter. More providers to come in future.
cost on metrics for runs on OpenRouter. More providers to come in future.@tool in its methods by @kausmeows in https://github.com/agno-agi/agno/pull/5701Full Changelog: https://github.com/agno-agi/agno/compare/v2.2.14...v2.3.15
Reasoning Streaming: Add support to stream reasoning chunks for the case where reasoning_model is provided.
reasoning_model is provided.output_schema now accepts JSON schemas in provider-specific formats, passed directly to the model API without transformation. This gives users full control over structured output configuration for OpenAI, Claude, and OpenAI-like providers.introduction messages.search_parameters to the search and async_search methods of the Milvus vector database class.reload_includes and reload_excludes to the serve function of AgentOS. Using these you can specify which files the AgentOS app listens to to reload.Full Changelog: https://github.com/agno-agi/agno/compare/v2.3.13...v2.2.14
secret_key has been deprecated (although still supported) and it is recommended to supply verification_keys=[...] going forward.
agents:my-agent:read you can control which agents can be run (i.e. when calling POST /agents/my-agent/runs ) or which agents are returned on read (i.e. when calling GET /agents or GET /agents/{id}). This allows for full control over which users can use and see which agents, and this extends to AgentOS. The same also applies to Team and Workflow.JWTMiddleware class:
secret_key has been deprecated (although still supported) and it is recommended to supply verification_keys=[...] going forward.algorithm default value changed from HS256 to RS256 to make asymmetric keys the default (better suited for production cases).Full Changelog: https://github.com/agno-agi/agno/compare/v2.3.12...v2.3.13
One column per month.