Skip to content

n8n@2.49.0

Pre-release
Pre-release

Choose a tag to compare

@n8n-assistant n8n-assistant released this 17 Nov 09:40
b2ce425

2.49.0 (2026-11-17)

Bug Fixes

  • Accept custom OpenAI-compatible model IDs in v2 (#43550) (5c7815a)
  • Add context to tool-call errors (#42474) (71941da)
  • AI Agent Node: Keep the model's text before tool calls in agent history (#43559) (f932917)
  • AI Agent Node: Keep tool-call history separate for each input item (#43890) (2b65631)
  • AI Agent Node: Save every tool-calling round of a turn to chat memory (#43536) (c06252a)
  • AI Agent Node: Send PDFs to AWS Bedrock with a text block and a valid document name (#43351) (6a70ee0)
  • AI Agent Node: Stop adding newlines between streamed text chunks (#43252) (d40d5ea)
  • ai-builder: Attach Gateway credits only to nodes that can use them (#43300) (7b6dc99)
  • ai-builder: Check per-action required fields in Assistant multi-action tools (#43303) (efc5fa3)
  • ai-builder: Correct the Data Tables node type and UTF-8 decoding in workflow best practices (#43278) (652c028)
  • ai-builder: Link existing resources in Assistant replies (#43375) (abe4379)
  • ai-builder: Make "Always allow" work on MCP approval cards in the n8n Assistant (#42913) (73c7791)
  • ai-builder: Reject workflow fields for Agent credential setup (#43262) (9909649)
  • ai-builder: Remove the preview disclaimer from n8n Assistant (#43384) (9255660)
  • ai-builder: Search service connections without an extra permission question (#43333) (6f8d6c0)
  • ai-builder: Show a taken app namespace on the plan card (#43267) (3997f81)
  • ai-builder: Use medium reasoning effort for GPT-5.6 and newer models in the Assistant (#43529) (d965df5)
  • Align credential display name with node branding (#44160) (b509ff1)
  • Allow reasoning effort for models on custom OpenAI-compatible endpoints (#43450) (dba20dd)
  • API: Reject edits to read-only credential nodes in the Public API (#42764) (1550a6f)
  • API: Require only the API key scopes that a package import uses (#43293) (e5b9b9b)
  • API: Support personal alias on public folder routes (#43230) (9f48ac9)
  • Azure OpenAI Chat Model Node: Record provider token usage in Responses mode (#43891) (ac9bfd3)
  • Broken e2e test in source-control-helper (#43816) (bec390b)
  • Check log streaming destination credential roles (#43622) (e11a259)
  • Chroma Vector Store Node: Require a Chroma Cloud API key (#43643) (bfce878)
  • Clarify Ignore Bots rejection message (#41064) (e4ef29f)
  • Cohere Chat Model Node: Preserve tool call history (#39523) (fefeebc)
  • core, editor: Show the specific data table and tag conflict reason in Apply (#43447) (69bbcb4)
  • core: Add trace headers to authenticated HTTP requests (#43791) (ca617d3)
  • core: Allow Assistant edits around unchanged Header Auth nodes (#43208) (4b417db)
  • core: Avoid loading workflow data once per tag and share (#41894) (00d090c)
  • core: Check briefly on a sandbox that failed to start earlier in the turn (#43699) (479f689)
  • core: Clear cached OIDC configuration when discovery fails (#43075) (a71a116)
  • core: Compare credential allowlist domains in their ASCII form (#43291) (de4c052)
  • core: Do not send promptCacheRetention to GPT-6 and later models (#44000) (7d75ca2)
  • core: Explain restart interruptions in Assistant and Agents messages (#43251) (decbc93)
  • core: Fail a migration when an index operation fails (#43615) (bed0094)
  • core: Fix tool call, abort signal and bound option handling in the LangChain chat model adapter (#43721) (dfdf594)
  • core: Give correct replacement advice for the Convert to/from binary data node (#43620) (d82e954)
  • core: Give multi-input nodes the same inputs in partial and full executions (#42726) (4207e89)
  • core: Hand dollar property reads to the expression engine (#43423) (ed626a7)
  • core: Improve error messages for when whatsapp fails to connect … (#42894) (86b9a96)
  • core: Keep durable trigger jobs when a publish follows an unpublish (#43821) (8a60e91)
  • core: Keep expression transfers working after built-in replacements (#43231) (dae98c2)
  • core: Keep nodes of unloaded types when the MCP server updates a workflow (#43621) (a190aa4)
  • core: Keep the observational memory window cap in agent configs (#43919) (53c9b48)
  • core: Keep the offloaded tool result marker true during its agent run (#43306) (519c7ed)
  • core: Keep tool-call IDs from OpenAI-compatible servers in agent history (#43328) (64b1fc7)
  • core: Keep trusted-source cache keys in one Redis cluster slot (#43599) (f82deca)
  • core: Let agents turn off prompt caching for endpoints that reject the cache fields (#43651) (b6b973b)
  • core: Limit Assistant credential suggestions to owned, shared, and global credentials (#43175) (a4f4db6)
  • core: Make Agents' OpenAI Responses calls stateless and recover from reasoning of another org (#43493) (497a0f5)
  • core: Match bare node references across expression engines (#43229) (a598c26)
  • core: Never produce an empty tool name after truncation (#43286) (6266bcd)
  • core: Preserve recovered execution data on stalled-job crashes (#40166) (8f43cfc)
  • core: Prevent agent wake replay after execution recording failures (#43226) (c374dfa)
  • core: Prune workflow history in batches outside the SQLite write lock (#43398) (b955cd3)
  • core: Put workflow custom tags on node spans (#43656) (8d277e5)
  • core: Reactivate inactive Daytona snapshots during sandbox creation (#36791) (779738a)
  • core: Read legacy-format data with the instance key when the stored key differs (#43319) (c2caa11)
  • core: Redact consumed auth headers case-insensitively (#43061) (328c09c)
  • core: Refresh Gateway credits tokens in long-lived agent runtimes (#41491) (0026cc1)
  • core: Remove stale project memberships when SSO project roles are empty (#43350) (8ef8121)
  • core: Report trigger connection failures from the user's setup as warnings (#43302) (c5c5be4)
  • core: Resolve External Secrets in log streaming webhook auth (#35449) (a0fee8f)
  • core: Return get_agent_session pages of up to 40,000 characters (#43843) (f8b49a0)
  • core: Return home project from workflow endpoint when sharing is disabled (#43353) (5f235c7)
  • core: Return not found for invalid waiting webhook execution ids (#43478) (18dedfe)
  • core: Run tool calls from Gemini's OpenAI-compatible endpoint (#43234) (2b8e937)
  • core: Sanitise errors from external vector stores (#43746) (a76176b)
  • core: Send agent channel action text as Markdown (#43921) (251c44b)
  • core: Send memory guard stop event with workflow owner as user (#43366) (65ec56b)
  • core: Send OTLP HTTP trace exports through the proxy from the environment (#43164) (3ab166c)
  • core: Settle child job waits on workers from completion events (#43142) (5ac388b)
  • core: Show no tag changes in the Apply preview when workflow tags are disabled (#43449) (bf20f97)
  • core: Stop flagging single-output vector store nodes in the Always Output Data v3 rule (#43619) (20827c0)
  • core: Stop redacting executions that use global credentials (#43223) (a2c518e)
  • core: Teach the n8n Assistant the OpenAI image analyze output shape (#41059) (04b9cfe)
  • core: Use resume attempt limit for failed resumes (#43641) (aa396aa)
  • core: Warn when a custom telemetry tag matches a built-in key (#43657) (077dff6)
  • Correct credential display name (#44161) (9d9f691)
  • Detect MCP JSON-RPC clients (#42855) (0000484)
  • Edit Image Node: Register font once instead of on every text operation (#43337) (47ceba8)
  • editor: Add paste option to the empty canvas context menu (#43212) (5cbd14f)
  • editor: Align Agent version naming and history actions (#43413) (6c566c2)
  • editor: Align animated node border corners (#42888) (5413bef)
  • editor: Allow moving workflows to team projects without the Sharing license (#43991) (0a4a1a2)
  • editor: Defer Agent Assistant chat creation until first message (#43702) (fd889ba)
  • editor: Filter the Assistant project dropdown by the search term (#43222) (7003ade)
  • editor: Fix ApprovalCard text overflow and row alignment (#43159) (7ed3a18)
  • editor: Hide queued message load errors in agent chat (#43558) (198b831)
  • editor: Initialize chat WebSocket when reposting a message (#43593) (3effaf2)
  • editor: Keep a new agent as a draft after returning from Settings (#43371) (e4e5c66)
  • editor: Keep agent tabs available for new drafts (#43421) (c2d7831)
  • editor: Keep the chat draft when Fix with AI sends its prompt (#42876) (d24a35c)
  • editor: Keep the inline agent model picker's credentials in the node settings (#43261) (63f82e2)
  • editor: Let every user close dynamic banners with permanent close (#43595) (f73c76a)
  • editor: Load older Assistant messages when you scroll up (#43321) (1f3a0fe)
  • editor: Localize the dismiss tooltip on callout close icons (#42588) (f216f7f)
  • editor: Localize the metric dropdown placeholder in the evaluations chart (#44068) (ad1de71)
  • editor: Localize the run at column in the evaluation runs table (#42823) (1ba982c)
  • editor: Localize the tabs and table header in the source control push and pull modals (#44079) (45fbeeb)
  • editor: Localize the workflows table in the community package modal (#42821) (8fd9f3e)
  • editor: Make JSON highlight color visible on light theme (#43652) (2e0e1ef)
  • editor: Make Web Search button fit its content (#43483) (d174960)
  • editor: Match Callout Markdown to small text size (#43543) (c0d6d7b)
  • editor: Open the Assistant logs panel only when a preview run fails (#43994) (795c59a)
  • editor: Preserve Assistant drafts during approvals (#43209) (034849c)
  • editor: Reduce Markdown H1 font size (#43482) (2967770)
  • editor: Refetch sub-workflow policy check when selecting a sub-workflow in the node panel (#42509) (d76d304)
  • editor: Resolve node placement size from rendered ports (#42777) (44111c1)
  • editor: Respect installation paths in Assistant artifact links (#43210) (18a63f9)
  • editor: Shorten the Teams setup copy for approving access (#42903) (0951d8d)
  • editor: Show a loading state on the workflow history Actions button (#43174) (8305cd0)
  • editor: Show Apps in the personal space permissions on the roles… (#43495) (e09e7d4)
  • editor: Show expanded canvas groups above sticky notes outside the group (#43390) (d81c2b6)
  • editor: Show pin indicators for workflow pin data after an execution (#43676) (d919f0f)
  • editor: Show the data table link when the value has no cached URL (#40837) (c004442)
  • editor: Show the full empty state in a narrow agent sessions list (#43370) (310541c)
  • editor: Show the selected owner in the Overview owner filter (#43893) (bd2ce42)
  • editor: Stop evaluation metric cards overflowing into their neighbours (#42155) (83efca2)
  • editor: Stop showing a stale execution error on canvas nodes (#43387) (76e9117)
  • editor: Stop the listbox scrolling the page on mount (#43311) (22f73a9)
  • editor: Use brand pink for active switch fill (#43806) (464d13b)
  • editor: Use the same localized toast when stopping a test webhook fails (#44074) (2d08d1d)
  • editor: Widen the migration report notify owner column (#43424) (4991c81)
  • Enforce Schedule Trigger interval limits in MCP (#43793) (765740d)
  • Fix v2 contact custom field updates (#41111) (8c6872d)
  • Google Gemini Chat Model Node: Show the block reason when Gemini returns no response (#43467) (dc3eeeb)
  • GoogleGemini Node: Use user role for tool results (#41021) (3333e4a)
  • Jira Trigger Node: Verify signed token on incoming Cloud (OAuth2) webhooks (#42800) (728360c)
  • Keep future-dated emails from advancing the poll cursor (#42476) (80d4344)
  • Limit credential checks and accept permission responses (#41119) (20b9cdd)
  • Lint resource operations in node classes outside .node.ts files (#39588) (7a702cd)
  • Make the programmatic example node template pass lint (#43492) (3502a57)
  • Make the scopes available label toggle the scope list (#41121) (eb858d0)
  • Microsoft Teams Trigger Node: Emit every change notification of a batched request (#43700) (ced436d)
  • Normalize API base URL paths (#41838) (611ede9)
  • Only warn for undefined match-column values (#41062) (1837541)
  • OpenAI Chat Model Node: Record provider token usage in Responses mode (#43129) (91d943c)
  • Pinecone Vector Store Node: Search indexes with a document schema through the documents API (#43451) (adecf5b)
  • Polish synced skill modal (#43508) (e8f505b)
  • Postgres Node: Yield to the event loop during item processing (#43367) (9f7ada0)
  • Preserve agent tool context on retries (#43608) (09118f7)
  • Preserve automatic API routing on custom endpoints (#43646) (62273bd)
  • Preserve canonical AI tool names in Agent V3 history (#39071) (d93033f)
  • Preserve non-UTF-8 bytes in asymmetric decryption (#41061) (13810f7)
  • Preserve streamed tool calls (#43546) (22d1a8a)
  • Qdrant Vector Store Node: Preserve reverse-proxy subpath as Qdrant client prefix (#43870) (58403b6)
  • Recognize separated credential field names (#41984) (8ad5e3d)
  • Refresh workflow project cache on project deletion (#43257) (78868f2)
  • Reject mismatched Gemma model responses (#41116) (a22cd8b)
  • Remove Duplicates Node: Prevent node from getting stuck when reaching maximum history size (#38494) (ff70691)
  • Remove New badges from Roles settings (#43859) (b70e354)
  • Require an API key in AI nodes that use API key credentials (#43560) (fb84d8b)
  • Require operation actions in community nodes (#39460) (76f66b4)
  • Reset string filter endpoints for each item (#41109) (21306b2)
  • Show the network cause of Gemini fetch errors and retry Google model requests less (#43625) (6b06660)
  • Stop creating agent versions when a followed skill is saved (#43318) (fc653a0)
  • Support Bearer Auth credentials (#43544) (533cd56)
  • Use Assistant credits in out-of-credits messages (#43748) (b10f99b)
  • Use dots spinner in buttons (#44008) (2f7278a)
  • Use only the full mail scope for service account tokens (#42475) (1f64042)
  • Weaviate Vector Store Node: Clear the collection only once per insert (#43518) (971dccd)
  • Weaviate Vector Store Node: Pass filters to hybrid search (#43517) (65f64bf)

Features

  • Add a webhook channel for agents (#42989) (114b71f)
  • Add CSP for apps on all responses (#43532) (5b80dbd)
  • Add skill source health and credential replacement (#43795) (a027ba9)
  • Add skills submenu to the Assistant input (#43312) (004dea2)
  • Add tool value limit and pages to get_agent_session MCP tool (#43794) (5e2b9fd)
  • ai-builder: Add opt-in observational memory policies for Instance AI (#43304) (a0cf9a5)
  • ai-builder: Add sample files for verification fixtures (#42951) (3db2fdc)
  • ai-builder: Add tool catalog mode to the agents SDK and use it in Instance AI (#43247) (34900ce)
  • ai-builder: Count Assistant verification helper calls in credits (#43499) (931f31c)
  • ai-builder: Give n8n Assistant step-specific tips with context cues (#43269) (d97fee4)
  • ai-builder: Run Assistant helper calls on a utility model (#43498) (4b21f80)
  • API: Rename the archived export flag to includeArchived (#43391) (34a0406)
  • API: Send an audit event when a Public API rate limit changes (#43712) (2b37d65)
  • API: Set the n8n version as meta-data in the OpenAPI spec (#43352) (50184df)
  • API: Show required API key scopes in operation descriptions (#43400) (f8d9fd0)
  • API: Use one change preview for the UI, the public API and the CLI (#43265) (fa918cf)
  • core, editor: Add agent channel descriptions and rename the Webhook channel to API (#43758) (d1c82f8)
  • core, editor: Add an Automatic OAuth mode that registers the client for selected services (#42782) (69e20f9)
  • core, editor: Show a toast when an imported workflow contains restricted nodes (#43553) (1b2def7)
  • core, editor: Support approval requests in the webhook agent channel (#43056) (0ab1817)
  • core: Add an archive state for agents (#43392) (b624a63)
  • core: Add audit events for promotion providers and environments (#43277) (fc39c94)
  • core: Add dangerous override functions to the otel.configure hook (#43729) (9f957e8)
  • core: Add dependencies tool to the MCP server (#43161) (c5cf0dc)
  • core: Add GenAI tool attributes to tool node spans (#43658) (1f86fc2)
  • core: Add migration report tools to the instance MCP server (#40817) (f18e048)
  • core: Add offset/limit pagination and workflow-parity filters to the agents lists (#43814) (1a080f7)
  • core: Agent config schema now can include tool guardrails (#43404) (3ae8f1a)
  • core: Align outbound MCP OAuth with the 2026-07-28 authorization spec (#42907) (9650387)
  • core: Carry the agent archive state in n8n packages (#43329) (075fb1f)
  • core: Carry workflow and credential descriptions in n8n packages (#43401) (f9dcb57)
  • core: Close promotion reviews from the Inbox (#43362) (9d04dd5)
  • core: Emit an audit event when a user reads credential data (#43063) (c11c7c7)
  • core: Explain common Git failures in Promotions (#43453) (2f2f34f)
  • core: Keep unattended task approvals in new status waiting (#41399) (44832d3)
  • core: POST public endpoint for agents (#43719) (3716c17)
  • core: Promote and apply projects and folders without selecting a workflow (#43260) (b52211f)
  • core: Public API DELETE endpoint for agents (#43562) (a44a347)
  • core: Replace the MCP agent delete tool with an archive tool (#43393) (d5954c8)
  • core: Run pattern-based tool guardrails on agent tool calls (#43664) (c2a518b)
  • core: Send audit event when unpublishing workflow automatically (#43162) (59dcfb5)
  • core: Support reading agents via GET in the Public API (#44011) (1d2eed1)
  • core: Support tags on agents in the internal REST API (#43324) (ece4c61)
  • core: Tell MCP clients how to save a preference when none are saved (#43645) (1c5e6a5)
  • core: Track executions that use personal credentials (#43246) (34f38e6)
  • editor: Add a shared settings enable row (#42471) (f853191)
  • editor: Add archive and restore for agents (#43394) (ca6482e)
  • editor: Add brand spinner for loading states (#43468) (9ebd46a)
  • editor: Add Upload folder and Upload SKILL.md to the Create skills menu (#43171) (63d480a)
  • editor: Clarify agent budget coverage with a dismissible caption (#43385) (2e7e96c)
  • editor: Cleans up nudge assistant on workflow error experiment logic and rolls out feature to all instances (#43435) (7d243d8)
  • editor: Move the user's links into an account menu at the bottom of the sidebar (#42642) (86162be)
  • editor: Show Gateway credits promotions from the Gateway config (#43475) (a4a7091)
  • editor: Show restricted template nodes on the template page and in the setup panel (#43739) (f75a553)
  • Enforce one active credential per community node (#39122) (32cb68b)
  • GitHub Node: Add Secret resource (#25441) (2082036)
  • Implements v3 iteration of existing n8n Assistant taxonomy experiment (#43692) (99163e2)
  • Let the publish_workflow MCP tool set the version name and (#43767) (952d2e7)
  • Let users edit and save skills in the Assistant artifact panel (#43314) (950b29c)
  • MCP Client Tool Node: Accept generic OAuth2 API credentials (#37371) (4c0ef3c)
  • Publish attach resources with the App (#43660) (3bf5fbe)
  • Scan a local community package directory before submission (#42948) (1f0aaa7)

Performance Improvements

  • ai-builder: Run time-based Wait nodes without their delay in verification runs (#42950) (bba21d5)
  • core, editor: Include public settings in the editor page (#43235) (bc9ded0)
  • core: Keep the replaceSpecialChars character map out of the runtime bundle (#43418) (eb27bae)
  • core: Reduce per-item setup in the expression runtime (#43335) (b327e2d)
  • core: Serve the QuickJS WASM file with a long-lived cache (#44023) (32865f9)
  • core: Ship only the release-sync QuickJS variant (#43363) (cfea9b9)
  • editor: Build the schema view without merging through reactive proxies (#42939) (cc30027)
  • editor: Lazy-load modal components to shrink the startup bundle (#43331) (547b8c8)
  • editor: Let the editor build drop unused n8n-workflow and api-types modules (#44006) (dd4e385)
  • editor: Reduce canvas status animation CPU usage (#43667) (03c895b)

View guided diff Turn on auto-fix