Self-host agno

Changelog

New Features:

• Native Approvals for Harness Agents: Claude's and Codex's own permission requests now reach the AgentOS approval UI. When the agent has a database, including one AgentOS supplies, a native request persists a PAUSED run that shows up in the existing chat and Approvals views, and Resolve plus Continue Run delivers the decision back to the same native execution without replaying the prompt. The harness's own policy decides when to ask — Claude keeps its permission_mode and explicit permission callbacks, Codex supports approval_policy="on-request" | "untrusted" | "never" alongside its existing SDK auto_review and deny_all modes — so there is no separate Agno setting to turn on. Decisions are validated against the exact action requested, with sync and async continuation, cancellation, expiry and storage-failure handling, and sandboxed agents keep this across Docker and AgentCore: capability negotiation, a decision RPC, fenced persistence, reattachment after an API-worker loss, separate approval-wait and execution budgets, and no replay after process loss. Note that a CodexAgent with no approval settings now runs under the native on-request policy, so a request that previously went nowhere pauses the run instead; hitl_timeout (default 900s) bounds how long it waits. See cookbook/frameworks/human_in_the_loop/.
• Initially Disabled Company Routines: Routine(enabled=False) lets a template ship an inactive Routine that a user turns on later from the UI, with True still the default. Disabled Routines are created atomically, stay linked to their Product and visible in listings, are skipped by the scheduler, and keep the user's saved choice across restarts. The flag is honored in Python and YAML bootstrap, the setup and creation APIs, the sync and async Company Manager, and the Company tools. Duplicate create requests still return 409 and cannot enable an existing Routine, and the existing pause/resume endpoints enable it without recreating it. See cookbook/05_agent_os/30_company/11_disabled_routine.py.
• Headline Numbers on Charts: ChartTools.create_chart() takes an optional preformatted value — a total or latest value such as "$22.3K" or "94%" — emitted as "value" in the chart block and rendered by Agno OS beside the chart title. The docstring tells the model to leave it out when no single number summarizes the chart, so averages and rates are not summed.

Improvements:

• Company Routine Sorting: GET /company/routines now also sorts by next_run_at, schedule_type and enabled, so Agno OS can sort its routines table by Next run at, Schedule and Active. Paused and on-demand Routines have no upcoming run, so they come last in either direction, and Routines with the same value keep their creation order.
• Cookbooks: Added an ApifyTools job search example that caps the Actor's item count with a tool hook and trims each dataset item to title, company, location, posted date and link, so a broad search stays small in context. Added a fully local Pixeltable RAG example that writes chunks through langchain-pixeltable and searches them with LangChainVectorDb, using local sentence-transformers embeddings and Ollama — no embedding API key and no separate vector database server.

Bug Fixes:

• FastAPI Native Tracing: FastAPI 0.142.0 enables its own OpenTelemetry tracing as soon as a global tracer provider exists, which AgentOS(tracing=True) installs, so FastAPI's request spans were stored alongside Agno's. Every run became a child of the request span and traces were recorded as fastapi.dependencies with empty run_id, session_id and agent_id, breaking the Runs view, while each UI request stored a trace of its own. AgentOS now opts the app it builds out of FastAPI's tracing, and setup_tracing() filters foreign spans so a stored trace is rooted at the Agno run. Apps without tracing=True keep FastAPI's defaults, and users' own spans created inside tools are still kept.
• Company Routine Cancel: POST /company/routines/{id}/cancel recorded the run as cancelled as soon as the executor accepted the request, but the agent only stops at its next checkpoint. For that window the API reported cancelled while the agent was still working, and archive, delete and executor change went through mid-run. The route now records a status only from durable evidence: a live run is left alone for ScheduleExecutor to settle, a finished run is written through the scheduler's own mapping, a missing run is recorded cancelled, and a crashed run past its timeout no longer blocks archive, delete or executor change.
• AccuracyEval Judge: AccuracyEval with no model and no evaluator_agent built its judge with OpenAIChat(id="o4-mini"), which OpenAI shut down on 2026-10-23, so every judge call returned 404 model_not_found, no iteration scored, and the eval reported an empty run instead of the provider error behind it. The default is now gpt-5.6-terra, the replacement OpenAI lists for o4-mini. The examples that pinned retired models move across too, and the accuracy examples switch to OpenAIResponses because OpenAIChat rejects function tools for these models.
• Knowledge Per-Upload Chunking: POST /knowledge/content accepts per-upload chunker, chunk_size and chunk_overlap, and process_content() applied them straight to content.reader — the shared custom or already-resolved reader on the knowledge base. Every later upload using that reader kept the previous upload's chunker until the process restarted. The upload now gets its own shallow copy of the reader before the chunking strategy is applied, so the knowledge base keeps its own configuration.

What's Changed

• fix: knowledge keep per-upload chunker options off the knowledge base reader by @sunhouy in #11327
• fix: move the AccuracyEval judge and examples off retired OpenAI models by @kausmeows in #11347
• [cookbook] Add a local Pixeltable RAG example by @pierrebrunelle in #11314
• [cookbook] ApifyTools job search with capped runs and trimmed results by @johnisanerd in #11305
• feat: show a headline number on ChartTools charts by @anuragts in #11110
• fix: keep FastAPI native spans out of Agno traces by @kausmeows in #11346
• fix: record a Routine cancel only once its run has stopped [SDK-643] by @harshsinha03 in #11352
• chore: README by @ashpreetbedi in #11380
• feat: support initially disabled Company routines by @ashpreetbedi in #11372
• feat: route native Claude and Codex approvals through AgentOS by @ashpreetbedi in #11374
• feat: sort routines by next run, schedule type and enabled by @Ayush0054 in #11382
• chore: core release v3.1.12 by @kausmeows in #11383

New Contributors

• @sunhouy made their first contribution in #11327
• @pierrebrunelle made their first contribution in #11314
• @johnisanerd made their first contribution in #11305

Full Changelog: v3.1.11...v3.1.12

Changelog

New Features:

• Harness Agents in Teams and Workflows: ClaudeAgent and CodexAgent can now be members of an Agno Team and steps of a Workflow, including inside Parallel, Loop, Condition and Router. The team leader delegates to them with delegate_task_to_member like any other member, each harness runs with its own model, tools and workspace, keeps its own conversation across turns of the same session, and the team or workflow collects its runs as usual. A harness agent with sandbox= runs each turn in its own container as a member or step, and AgentOS lists harness members on the team page. See cookbook/frameworks/harness_teams/ and cookbook/frameworks/harness_workflows/.
• Session Search: Added GET /sessions/filter-schema and POST /sessions/search with nested AND/OR/NOT filters on session name, session/user/component IDs, session type and created/updated timestamps, matching the Traces API. Predicates are applied before counting and pagination in PostgreSQL and SQLite, with a batched fallback for other adapters. Sync and async client methods are included.
• Company Tasks: POST /company/tasks accepts status, so Tasks can be created in Backlog, Ready or Cancelled. Tasks can move between any stages, Done and Cancelled Tasks can reopen, and title and brief can be edited at any stage. In progress still starts only with a run, and moving out of it while that run is executing or paused is refused.
• Company Routines: Added POST /company/routines/{id}/cancel, /archive, /restore and admin-only DELETE /company/routines/{id}. GET /company/routines takes sort_by and sort_order. Pause and resume on an on-demand Routine return 409, and scheduler runs no longer move updated_at.

Improvements:

• Cookbooks: Added a customer feedback workflow for the AgentOS UI, where Company Manager reads Markdown feedback, proposes evidence-backed Projects and Backlog Tasks, and leaves acceptance to a human.
• README: Aligned the README with the docs introduction and the self-improving agent platform positioning.

Bug Fixes:

• ReaderFactory: create_reader() now returns a fresh reader configured with the per-call options instead of a cached singleton, so chunk_size and other overrides are respected and mutations no longer leak across Knowledge instances.
• Workflow Step Content: StepOutput.to_dict() keeps ints, floats, bools and strs as themselves instead of stringifying them, so a step that returned 0 or False no longer comes back as "0" or "False" after a HITL resume, a session reload, or through the AgentOS workflow API.
• HuggingFace: default_headers is mapped to the SDK's headers, store is sent in extra_body, and max_retries and default_query are deprecated and ignored instead of raising TypeError. An agent with no tools no longer sends tools=[], which made non-streaming runs return None content.
• Google Calendar: find_available_slots now works in the authenticated user's timezone instead of UTC, so working hours and date-only inputs mean the user's local day. It also fetches events for the full scanned window, so busy time on the last day of a range is no longer offered as free.
• Company Archive: Archived Tasks no longer block archiving their Project or Product. Visible active Tasks still block as before.
• RedisDb: keyword_search now passes the caller's limit to redisvl instead of silently returning its default of 10 results.
• MMRReranker: rerank and arerank return no documents when limit is zero or negative instead of always returning the closest match.
• FieldLabeledCSVReader: page_size=0 is rejected alongside negative values instead of silently returning no documents on larger files.
• AgentOSClient: Removed the documented but non-existent stream argument from continue_agent_run. Use continue_agent_run_stream to stream.

What's Changed

• cookbook: add customer feedback UI workflow by @ashpreetbedi in #11271
• chore: refresh platform README by @ashpreetbedi in #11291
• chore: clarify README self-improvement workflow by @ashpreetbedi in #11293
• [fix] Return no documents when the MMR reranker limit is zero by @makiaveli1 in #10450
• fix: drop the non-existent stream arg from continue_agent_run by @yetuge in #10453
• chore: align README with self-improving agent platform positioning by @ashpreetbedi in #11301
• fix: pass the caller's limit to redisvl in RedisDb.keyword_search by @Ayoubhm07 in #10781
• fix: don't let archived Tasks block Company archive by @Ayush0054 in #11292
• fix: query the scanned window in find_available_slots by @devYRPauli in #10706
• fix: keep JSON-native step content types across a workflow session reload by @kausmeows in #11304
• fix: send HuggingFace client and request params under the names the SDK takes by @kausmeows in #11307
• [fix] Reject zero page size in field-labeled CSV reader by @jiangjiang248 in #10602
• chore: align README with docs introduction by @ashpreetbedi in #11313
• fix: use the authenticated user's timezone for calendar slots by @Himanshu040604 in #11306
• fix: ensure ReaderFactory respects per-call options and avoids mutable shared state by @Fazlul105 in #11296
• test: move the OpenAI reasoning tests off the retired o3-mini by @kausmeows in #11331
• test: move the team event streaming tests off the retired o3-mini by @harshsinha03 in #11335
• feat: let Claude Code and Codex agents be members of an Agno team by @kausmeows in #11241
• feat: add advanced session search and filter schema by @pritipsingh in #11319
• feat: Company Task stages on create and move, Routine cancel, archive and delete by @harshsinha03 in #11312
• fix: don't let archived Tasks block Project completion or retention by @ysolanky in #11336
• chore: Release v3.1.11 by @kausmeows in #11334

New Contributors

• @Ayoubhm07 made their first contribution in #10781

Full Changelog: v3.1.10...v3.1.11

Changelog

Bug Fixes:

• CodexAgent: Now signs in automatically with OPENAI_API_KEY, and keeps an existing ChatGPT or other login.
• OS Manager: OS Manager now lists and runs external agents such as ClaudeAgent and CodexAgent registered in AgentOS(agents=[...]), by name or ID.
• Knowledge Upload: POST /knowledge/content now returns a clear 400 message when an upload has no file, text_content or URL, or when url is not a single URL.
• Tracing: Spans from AgentOS(tracing=True) and setup_tracing() now keep up to 1,024 attributes, so model spans keep their full input and message history on runs with many tool calls. OTEL_SPAN_ATTRIBUTE_COUNT_LIMIT and OTEL_ATTRIBUTE_COUNT_LIMIT still take precedence.

What's Changed

• chore: log replica-local queue stream notice at info level by @ysolanky in #11264
• fix: reword the no-authorization user isolation warning by @SamJupe in #11263
• fix: authenticate Codex automatically with OPENAI_API_KEY by @ashpreetbedi in #11249
• fix: let OS Manager discover and run external agents by @ashpreetbedi in #11247
• fix: clarify accepted Company Task execution by @ashpreetbedi in #11261
• fix: reject knowledge uploads without a content source by @ashpreetbedi in #11259
• fix: clarify readable Company sources for agents by @ashpreetbedi in #11256
• fix: reject unsupported knowledge URL input types by @ashpreetbedi in #11260
• fix: use LLM-appropriate tracing attribute limits by @ashpreetbedi in #11255
• fix: keep Company Task projection off the queue's critical path by @ysolanky in #11188
• fix: allow authorized members to trigger Company routines by @ashpreetbedi in #11254
• feat: govern Company writes by scope instead of ownership by @SamJupe in #11270
• chore: Release v3.1.10 by @harshsinha03 in #11265

Full Changelog: v3.1.9...v3.1.10

Changelog

New Features:

• MCP Tool Runner: POST /mcp/server/tools/{tool_name}/run runs any tool listed on the AgentOS Server Card over HTTP and returns its result, so a tool can be tested without an MCP client. Turn it off with MCPConfig(tool_run_endpoint=False). See cookbook.

Bug Fixes:

• AgentOS MCP Server: /mcp and /mcp/server-card now accept browser requests from origins the AgentOS CORS policy allows, so os.agno.com can reach the MCP server of a local AgentOS. /info now reports the configured MCP path instead of always /mcp.
• Postgres Metrics: Metrics refresh on a migrated database no longer loads the legacy sessions.runs column into memory.
• Function Workflows: A function passed as a workflow's steps now runs once when its body raises a TypeError, and sync functions and generators now work with arun().
• AgenticChunking: Chunking no longer hangs when the model returns a zero or negative breakpoint, and max_chunk_size must be greater than 0.
• BaiduSearchTools: Chinese titles and abstracts are returned as readable text instead of \uXXXX escapes.
• Human-in-the-Loop User Input: A requires_user_input tool with a parameter typed int | None or another X | Y type now pauses cleanly, and the paused run is saved and resumes with continue_run(run_id=...).
• Mistral: The timeout, endpoint and safe_mode parameters on MistralChat, and endpoint on MistralEmbedder, now work instead of raising a TypeError. safe_mode is deprecated in favor of safe_prompt.

What's Changed

• [fix] Keep Chinese text readable in BaiduSearchTools output by @RaycarlLei in #11229
• fix: handle non-positive breakpoints and validate max_chunk_size in AgenticChunking by @Fazlul105 in #11231
• fix: run custom workflow steps function once when its body raises TypeError by @kausmeows in #11235
• fix: stop Postgres metrics refresh from loading the legacy runs column into memory by @sannya-singal in #11190
• feat: run Company automation hourly instead of every minute by @ysolanky in #11221
• fix: admit CORS-allowed browser origins on the MCP surface by @ysolanky in #11246
• feat: run a published MCP tool over HTTP by @sannya-singal in #10666
• fix: serialize PEP 604 tool parameter types in UserInputField.to_dict by @ShaoLOWu in #11177
• refactor: rename Product-scoped Company tools by @ashpreetbedi in #11242
• fix: map Mistral client params onto the mistralai SDK and alias safe_mode to safe_prompt by @ysolanky in #11222
• chore: Release v3.1.9 by @harshsinha03 in #11248

New Contributors

• @Fazlul105 made their first contribution in #11231
• @ShaoLOWu made their first contribution in #11177

Full Changelog: v3.1.8...v3.1.9

Changelog

New Features

• Postgres Sandbox Checkpoint Store: Added PostgresCheckpointStore for storing sandbox checkpoints directly in PostgreSQL, allowing multi-replica deployments to restore sessions without S3 or shared filesystems. Checkpoints are chunked, written transactionally, deduplicated by SHA-256 and verified on read. It is opt-in; existing checkpoint defaults are unchanged. (#11186)
• Framework Managers as Routine Executors: Company Routines can now use the internal Product Manager or OS Manager as executors through declarations, the Routine API and human-chat configuration. Reserved manager IDs resolve only from the owning AgentOS and cannot be replaced by saved or user-defined components. Includes an optional weekly Product Manager review. (#11193)
• Sandbox Checkpoint Retention: Sandboxes now retain the three newest checkpoints per session by default. Configure retain_checkpoints=None to preserve every checkpoint. Unpublished or superseded checkpoints are retired and deleted after the run outcome is safely stored. (#11187)

Improvements

• Sandbox Session Cleanup: Deleting a sandbox-backed session now removes its warm computer and associated checkpoint objects after the database transaction commits. Single, bulk and user-scoped session deletion are supported without affecting sessions the caller does not own. (#11187)
• OS Manager Specialist Visibility: Platform Builder and Platform Manager now appear as members of OS Manager in /teams and /teams/os-manager-internal. They remain excluded from the reusable application registry, top-level agent discovery and standalone agent routes. (#11198)
• Claude Defaults to Sonnet 5.5: Claude() now defaults to claude-sonnet-5-5, replacing the deprecated claude-sonnet-4-5-20250929 before its retirement. Non-default temperature, top_p and top_k values are dropped with a warning for newer Claude models that reject those sampling parameters. This behavior applies across Anthropic, Bedrock, Foundry and Vertex providers. (#11212)

Behavior Changes

• Checkpoint Retention: Only the three newest sandbox checkpoints are retained by default. Opening a checkpoint for an older run can now raise CheckpointObjectNotFound. Set retain_checkpoints=None to restore the previous unlimited-retention behavior. (#11187)
• Durable Sandbox Startup Validation: AgentOS now fails at startup when a durable-queue sandbox uses the host-local default checkpoint store. Configure a checkpoint store shared by every replica, such as PostgresCheckpointStore or S3CheckpointStore. Previously this invalid configuration failed only on the first submission. (#11187)

Bug Fixes

• Missing Sandbox Checkpoints: A committed checkpoint missing from storage now ends the turn with a clear, non-retryable error, retires the newly created computer and releases the session. Previously the run remained RUNNING, exhausted queue retries and blocked later turns on the session. (#11187)
• Company Routine Execution: Denied status writes and failed Studio runner delegation now correctly fail the Routine. Stale archived catalog entries no longer collide with owned reserved managers. (#11193)

What's Changed

• fix: sandbox checkpoint recovery, retention and session-delete cleanup by @ysolanky in #11187
• fix: show platform specialists under OS Manager by @ashpreetbedi in #11198
• feat: add PostgresCheckpointStore for sandbox checkpoints in Postgres by @ysolanky in #11186
• feat: support framework managers as Company Routine executors by @ashpreetbedi in #11193
• fix: default Claude to sonnet-5-5 before sonnet-4-5 retires by @kausmeows in #11212
• fix: default OpenAITools image generation to gpt-image-2 by @23821 in #10824
• chore: split Company db suites out of test-db and add a sixth unit-test shard by @harshsinha03 in #11213
• fix: send adaptive thinking when a Claude model rejects a thinking budget by @harshsinha03 in #11215
• chore: Release v3.1.8 by @kausmeows in #11208

New Contributors

• @23821 made their first contribution in #10824

Full Changelog: v3.1.7...v3.1.8