RecipeAgents / n8n nodes / AI Agent
Node reference · data pulled Oct 26, 2026n8n AI Agent node: what the data shows
Lets a chat model decide which connected tools to call, step by step, to finish a task.
What it does
The agent is a root node: on its own it does nothing. You attach a chat model, usually memory, and tools (other workflows, HTTP calls, vector stores, apps), and it loops between thinking and calling tools until it has an answer. Since n8n 1.82 every agent works as a tools agent; the old agent-type setting is deprecated.
It appeared in 4,246 public templates, but its share was lower in 2026 than in 2025 (see the year table). Template counts show use in public examples, not in production.
How often public templates use it
4,246 of the 13,574 public templates on n8n.io (31.3%) contain this node, in the snapshot we took on 2026-10-20. The table splits that by the year each template was published. 2026 runs to October 20.
| Year published | Templates with the node | Share of that year |
|---|---|---|
| 2024 | 68 of 462 | 14.7% |
| 2025 | 2,493 of 6,564 | 38.0% |
| 2026 | 1,675 of 5,845 | 28.7% |
The median template using this node has 406 views on n8n.io, against 284 for all templates. Template counts show what people publish, not what runs in production.
What it is used with
The most frequent partners are the nodes that appear most often in the same templates. "Lift" compares that with the whole library: 2× means the partner turns up twice as often next to this node as in an average template.
| Most frequent partner | Share of its templates | Lift |
|---|---|---|
| Edit Fields (Set) | 59.4% | 1.07× |
| OpenAI Chat Model | 57.9% | 2.53× |
| Code | 56.0% | 0.94× |
| If | 46.4% | 0.92× |
| HTTP Request | 45.4% | 0.88× |
| Structured Output Parser | 38.5% | 2.40× |
Strong pairings (in at least 5% of its templates, lift 1.5× or more): 13. The top ones:
| Strong partner | Share of its templates | Lift |
|---|---|---|
| Simple Memory | 25.3% | 3.16× |
| Call n8n Workflow Tool | 5.7% | 2.98× |
| Embeddings OpenAI | 6.9% | 2.67× |
| Chat Trigger | 14.9% | 2.66× |
| OpenAI Chat Model | 57.9% | 2.53× |
| Default Data Loader | 7.2% | 2.45× |
| Structured Output Parser | 38.5% | 2.40× |
| Anthropic Chat Model | 7.8% | 2.39× |
Most viewed public templates that use it: Ai agent chat (1,847,731 views); Build your first ai agent (1,256,812 views); Building your first whatsapp chatbot (541,254 views).
Versions and changes
Node versions from the npm packages n8n-nodes-base and @n8n/n8n-nodes-langchain, version 2.44.4. New workflows get the default version; workflows built earlier keep the version they were built with until you replace the node.
| Node | Versions | Default | Credential types | Scope |
|---|---|---|---|---|
| AI Agent | 1, 1.1, 1.2, 1.3, 1.4, 1.5, 1.6, 1.7, 1.8, 1.9, 2, 2.1, 2.2, 2.3, 3, 3.1 | 3.1 | none needed | – |
Source on GitHub: packages/@n8n/nodes-langchain/nodes/agents/Agent: 156 commits since 2023-11-29, last on 2026-10-22.
Recent fixes and features
- 2026-09-29: fix(Agent Node): Remove outdated tutorial tip callout (PR #35505)
- 2026-09-22: fix(AI Agent Node): Let an AI Agent Tool use its own tools under a pre-v3 parent agent (PR #39160)
- 2026-09-10: feat(AI Agent Node): Add option to force a tool call on the first iteration (PR #38342)
- 2026-09-07: fix(Anthropic Chat Model Node, AI Agent Node): Send disabled thinking to the provider and parse content-block output (PR #37954)
- 2026-08-04: fix(AI Agent Node): Stop pre-tool-call text leaking into agent responses (PR #35212)
- 2026-07-06: fix(AI Agent Node): Return intermediate steps when no tools are used (PR #33450)
Merged dates from GitHub. A fix only helps once your n8n instance runs a release that includes it, so check your version.
Known problems
- A chat model must be connected. The agent won't run without a chat model sub-node attached. Source: n8n docs
- No prompt specified. When the input isn't from a chat trigger, switch the prompt source to define it yourself and point it at the right field. Source: n8n docs
- 400 Invalid value for 'content'. The prompt resolved to null, often an expression pointing at a field that doesn't exist on the item. Source: n8n docs
- Text leaking before tool calls. An August 2026 fix stopped the model's pre-tool-call text appearing in the agent's final reply. Source: n8n PR #35212
- v1 will be removed. Agent node v1 (with the agent-type setting) is due to be removed in n8n 3.0; update old templates. Source: n8n docs
Forum trouble signal: not measured. The n8n community forum's terms forbid automated access, so we don't count forum threads.
Our recipes that use it
1 of our 25 importable workflows uses this node:
Official docs and sources
Usage, pairings, AI share and views: our count of 13,574 public n8n.io templates (snapshot 2026-10-20). Versions: npm packages, version 2.44.4. Changes: n8n's GitHub repository. Docs read and data pulled Oct 26, 2026. We describe each node in our own words; the docs are the authority on settings.
Related nodes
- Simple Memory1,086 templates · score 71.7
- OpenAI Chat Model3,104 templates · score 69.6
- Structured Output Parser2,181 templates · score 68.3
- Google Gemini Chat Model1,285 templates · score 65.1
- All 22 nodesSortable table with the scoring method
Questions
- How many n8n templates use the AI Agent node?
- 4,246 of the 13,574 public templates on n8n.io (31.3%), in our snapshot of 2026-10-20.
- What is the AI Agent node most often used with?
- Edit Fields (Set), which appears in 59.4% of its templates. The most distinctive partner is Simple Memory (lift 3.2×).
- Does the n8n AI Agent need a tool?
- n8n's docs say to connect at least one tool. Without tools, a Basic LLM Chain node is simpler.
- Which model do most n8n agents use?
- In the public library, the OpenAI Chat Model is attached in most agent templates; see the partner table for the exact share.