- 5 min read
- 8 key claims
- 5 sources cited
- For Developers, AI makers, Freelancers, Businesses
Overview
The n8n-io/n8n repository published version 2.48.5 on 2026-11-18 from commit 71592af, which carries a GitHub-verified signature [1][2]. The release notes for this version compare it against 2.48.4 and list the changes under bug fixes [1][2]. The stable release entry pointed to the same commit and version [1].
n8n describes itself as a fair-code workflow automation platform with native AI capabilities, combining visual building with custom code, available self-hosted or in the cloud, with more than 400 integrations [1][2]. That description and the integration count come from the project's own release pages and are not independently confirmed.
The patch is a maintenance release rather than a feature release. Two changes land directly on AI agent execution: tool-call history is kept separate for each input item, and newlines are no longer added between streamed text chunks [2].
Why it matters
The two AI Agent Node fixes address forms of interference in agent runs. Keeping tool-call history separate for each input item means context recorded while handling one item is not carried into another item in the same batch [2]. Stopping the insertion of newlines between streamed text chunks produces cleaner streamed output for downstream text handling [2]. Taken together, the changes read as a focus on predictable agent behavior in batched and streaming runs rather than on new capability.
Other fixes touch operational and access-control paths. Instance owners and admins are now treated as team project members in credential checks [2]. Workflows can be moved to team projects without the Sharing license [2]. The Features API now sends an audit event when a Public API rate limit changes [2]. The Qdrant Vector Store Node preserves a reverse-proxy subpath as the client prefix [2], and the Microsoft Teams Trigger Node emits every change notification within a batched request [2].
Opportunities
Teams processing many records through a single agent workflow can use the per-item tool-call isolation to make multi-item runs more predictable without restructuring the pipeline [2]. The streaming fix supports agent outputs that feed into document, chat, or record-writing steps without stripping stray line breaks [2].
For self-hosted deployments, the Qdrant fix reduces setup work for retrieval stacks that place the vector store under a path prefix behind a reverse proxy [2]. The Teams trigger change lets builders rely on each batched change notification being emitted individually when mirroring team activity into downstream systems [2]. The audit event on Public API rate-limit changes gives operators a record of a configuration change that affects API consumers [2].
Risks
A patch release can still change behavior in existing workflows. Routing dollar property reads to the expression engine and keeping small objects inside an Error in sandboxed expressions both alter how expressions evaluate and report failures, so workflows that depend on the previous handling may need rechecking [2].
Allowing workflows to move to team projects without the Sharing license removes a licensing gate, and organizations that use that license as a control may need to review their own permissions model [2]. The new rate-limit audit event adds entries to audit streams, which may affect log volume or downstream event processing [2].
The n8n Assistant taxonomy work is listed as a v3 iteration of an existing experiment [2], so building a production interface on it carries continuity risk. Because the relationship between the 1.x and 2.x lines is not documented, teams should confirm which line their deployment tracks before assuming a fix applies to it [3][4][5]. Upgrade content beyond 2.48.5 is also unassessable from the available notes, since 2.48.6 and 1.123.99 have no captured change list [4][5].
Context
n8n is a workflow automation platform that n8n itself describes as fair-code, with native AI capabilities and more than 400 integrations [1][2]. The repository publishes two version lines at the same time. Version 1.123.98 was published on 2026-11-18 from commit d2b5819 alongside the 2.x line [3]. The next day, 1.123.99 was published from commit 9e43bca [4], and 2.48.6 was published from commit 039c027 [5]. The captured 2.48.6 page was marked "Latest", as was the captured 2.48.5 page, but the later date on 2.48.6 indicates it supersedes 2.48.5.
The evidence does not state how the 1.x and 2.x lines differ in features, support policy, or intended deployments, and it contains no change list for 1.123.98, 1.123.99, or 2.48.6 [3][4][5].
Known, claimed, inferred
Known: 2.48.5 was published on 2026-11-18 from commit 71592af with a GitHub-verified signature, and its notes compare it against 2.48.4 [1][2]. The entries are grouped as bug fixes, with one item described as a v3 iteration of the n8n Assistant taxonomy experiment [2].
Hypothesis: each listed AI Agent Node change removes a form of state or output interference, so the release appears aimed at making agent steps behave more predictably in batched and streaming runs rather than adding features [2].
Hypothesis: publishing 1.123.x and 2.48.x on the same dates is consistent with maintaining a parallel release line, but the purpose of that line is not stated [3][4][5]. The later 2.48.6 and 1.123.99 releases, both dated 2026-11-19, supersede 2.48.5 and 1.123.98 respectively; the "Latest" label on the captured 2.48.5 page reflects when that page was recorded [2][4][5].
Limitations & uncertainties
No changelog entries were captured for 1.123.98, 1.123.99, or 2.48.6, so their content is unknown [3][4][5]. The evidence does not state how the 1.x and 2.x lines differ in features, support policy, or intended deployments [3][4]. There are no adoption, usage, or performance figures, and no independent evaluation of the AI Agent Node or Qdrant changes [2]. No published third-party test report is included, so no test results can be reported. The 400+ integrations figure and the platform description come from n8n's own release pages and are not independently confirmed [1][2].
Evidence & sources
The key statements of this analysis, labeled by type, with the sources they rely on. Trendoviz does not run product tests.
- Primary-source or vendor statementn8n 2.48.5 was published on 2026-11-18 from commit 71592af, which carries a GitHub-verified signature, and its notes compare it against 2.48.4.[1][2]
- Primary-source or vendor statementThe release is composed of bug fixes, including two AI Agent Node changes: keeping tool-call history separate for each input item and stopping newlines between streamed text chunks.[2]
- Primary-source or vendor statementCore changes give fresh installs a random instance ID, route dollar property reads to the expression engine, keep small objects inside an Error in sandboxed expressions, and treat instance owners and admins as team project members in credential checks.[2]
- Primary-source or vendor statementThe editor changes allow moving workflows to team projects without the Sharing license and match Callout Markdown to small text size.[2]
- Primary-source or vendor statementThe Qdrant Vector Store Node now preserves a reverse-proxy subpath as the client prefix, and the Microsoft Teams Trigger Node emits every change notification of a batched request.[2]
- Primary-source or vendor statementThe Features API sends an audit event when a Public API rate limit changes.[2]
- Primary-source or vendor statementn8n describes itself as a fair-code workflow automation platform with native AI capabilities and more than 400 integrations.[1][2]
- Primary-source or vendor statementVersions 1.123.98 and 2.48.5 were published on 2026-11-18, followed by 1.123.99 and 2.48.6 on 2026-11-19, with no change list captured for any of the three later or parallel entries.[3][4][5]
Sources
- 1n8n-io/n8n stable●github.comretrieved Nov 18, 2026
- 2n8n-io/n8n n8n@2.48.5●github.comretrieved Nov 18, 2026
- 3n8n-io/n8n n8n@1.123.98●github.comretrieved Nov 18, 2026
- 4n8n-io/n8n n8n@1.123.99●github.comretrieved Nov 19, 2026
- 5n8n-io/n8n n8n@2.48.6●github.comretrieved Nov 19, 2026
Updates & corrections
- First published · approved by an editor
Found an error? Report it.