n8n
by n8n GmbH
Workflow automation for technical teams. n8n is a workflow-automation and AI-orchestration platform that connects applications and services through a visual, node-based editor while allowing arbitrary code inside any step. It was founded in 2019 by Jan Oberhauser and is based in Berlin, Germany.
Performance Scores
22 rankings evaluated
Score range: 5.7 – 9.2
-
#1Most Customizable AI Agent Platforms 2026
Score: 8.1 · Best for: Technical teams wanting maximum control without giving up a visual canvas
-
#1Best Workflow Automation Tools 2026
Score: 8.7 · Best for: Technical teams wanting flexibility and self-hosting
-
#1Best Open Source Automation Platforms 2026
Score: 9.2 · Best for: Best for: technical teams wanting the most complete open-source automation platform
-
#1Best Free Automation Tools in 2026
Score: 8.5 · Best for: Unlimited self-hosted automation for technical teams
-
#1Best Automation Tools for Developers and Engineering Teams in 2026
Score: 9.0 · Best for: Developer teams wanting a self-hosted visual platform with code escape hatches
-
#1Best Automation Tools for Startups in 2026
Score: 8.5 · Best for: Technical founders wanting full control with unlimited free self-hosting
-
#1Best No-Code Automation Platforms in 2026
Score: 8.3 · Best for: Engineering-adjacent teams that want a visual builder with the option to self-host for cost or data residency.
-
#1Best Automation Tools by Price-Performance Value in 2026
Score: 9.2 · Best for: Teams with technical capacity wanting maximum automation per dollar
-
#2Best ETL & Data Pipeline Tools 2026
Score: 8.0 · Best for: Visual ETL pipelines with strong transformation nodes and broad connectivity
-
#3Best AI Agent Platforms in 2026
Score: 7.7 · Best for: Self-hosted AI workflow orchestration with code
-
#4Best Open-Source Workflow Engines for Engineers in 2026
Score: 7.9 · Best for: Mixed technical teams wanting a self-hosted Zapier alternative with both visual editing and scriptable code nodes
-
#5Best AI-Powered Automation Tools in 2026
Score: 7.2 · Best for: Developers building AI-augmented workflows with full infrastructure control
-
#5Best Customer Support Automation Tools in 2026
Score: 7.5 · Best for: Self-hosted support workflow automation
-
#5Best Human-in-the-Loop Automation Platforms 2026
Score: 6.7 · Best for: Technical teams running their own workflows, where the approver is a known colleague on a trusted channel
-
#6Best Durable Workflow Engines for Production in 2026
Score: 7.4 · Best for: Engineering teams that want visual workflow orchestration with a self-hosted, code-extensible core
-
#6Best Automation Tools for Retail & Ecommerce in 2026
Score: 7.3 · Best for: Self-hosted ecommerce data pipelines
-
#6Best Automation Tools for Data Teams in 2026
Score: 7.3 · Best for: Mixed API and data workflows with self-hosting
-
#6Best Process Orchestration Platforms 2026
Score: 6.9 · Best for: Mixed technical teams orchestrating across many SaaS systems
-
#6Best Automation Tools for Healthcare in 2026
Score: 5.7 · Best for: Healthcare IT teams with platform engineering capacity that need full-stack data residency
-
#7Best iPaaS and Integration Platforms 2026
Score: 7.5 · Best for: Developer teams wanting self-hosted automation with code flexibility
-
#7Best Automation Tools for Fintech and Financial Services in 2026
Score: 6.5 · Best for: Fintech startups and lenders with engineering capacity that need data to stay in-VPC
-
#7Best Automation Platforms for AI Orchestration 2026
Score: 7.1 · Best for: Technical teams exposing self-hosted workflows to an agent
Key Facts
Pricing
| Attribute | Value | As of | Source |
|---|---|---|---|
| Cloud pricing (EUR) | n8n Cloud (June 2026): Starter €20/mo (2,500 executions); Pro €50/mo (10,000 executions); Business €667/mo; Enterprise custom. Permanent free Cloud tier discontinued (trial only) | Jul 2026 | n8n Pricing page |
| n8n Cloud pricing (July 2026) | n8n Cloud (verified July 2026): Starter $24/mo (~$20 annual, 2,500 executions), Pro $60/mo (~$50 annual, 10,000 executions), new Business $800/mo (~$667 annual, 40,000 executions, 30 concurrent, self-hosted option), Enterprise custom. Self-hosted Community Edition remains free with unlimited workflows and executions. | Jul 2026 | n8n pricing |
| Execution billing model | Cloud billing is per workflow execution: one execution = one full workflow run regardless of node count or data volume (not per-step or per-task) | Jul 2026 | n8n Pricing page |
Technical
| Attribute | Value | As of | Source |
|---|---|---|---|
| License model | Source-available under the Sustainable Use License (fair-code): free to self-host for internal and personal use; reselling n8n as a hosted service requires authorization | Jul 2026 | n8n LICENSE.md (GitHub, master branch) |
Core Features
| Attribute | Value | As of | Source |
|---|---|---|---|
| Integrations | 500+ | Oct 2026 | n8n.io |
| Workflow Templates | 13,000+ | Oct 2026 | n8n Workflow templates library |
General
| Attribute | Value | As of | Source |
|---|---|---|---|
| Latest Funding | $180 million Series C at $2.5 billion valuation (October 2025). Total funding: $240 million | Jul 2026 | n8n blog — Series C announcement |
| Key Investors | Led by Accel; includes NVIDIA NVentures, Meritech, Redpoint, Sequoia, Felicis, Highland Europe | Jul 2026 | n8n blog — Series C announcement |
| GitHub Stars | 208,000+ | Oct 2026 | GitHub API |
| Deployment | Cloud, Self-hosted, Docker | Sep 2026 | n8n Docs — Choose how to use n8n |
| Users | 6x user growth and 10x revenue growth in the year to October 2025 (n8n-published; n8n publishes no absolute user count) | Jul 2026 | n8n blog — Series C announcement |
| Free Tier | Community Edition (self-hosted) | Oct 2026 | n8n Docs — Self-hosted community edition features |
| ARR | Beyond $40 million ARR as of October 2025 | Jul 2026 | Sacra |
| Cloud Pricing | Starter: EUR 24/month billed monthly or EUR 20/month billed annually (2,500 executions); Pro: EUR 60/month or EUR 50/month billed annually (10,000 executions). Business (EUR 800/month or EUR 667/month billed annually, 40,000 executions, SSO and LDAP) is available for self-hosted n8n only, with a hosted version on a waitlist (as of October 2026) | Oct 2026 | n8n Pricing page |
| Startup Program | EUR 333/month billed annually for companies with fewer than 20 employees and less than EUR 5M in total funding. 40,000 executions, Business-plan features | Oct 2026 | n8n Start-up Plan page |
| Latest Version | 2.x | Oct 2026 | n8n GitHub Releases API |
| US Market | Approximately 50% of global user base is US-based as of October 2025 | Jul 2026 | TechCrunch |
Support
| Attribute | Value | As of | Source |
|---|---|---|---|
| Community Size | 200,000+ members | Oct 2026 | n8n homepage |
AI Features
| Attribute | Value | As of | Source |
|---|---|---|---|
| AI Features | AI agent capabilities for building autonomous reasoning workflows, LangChain integration, and vector store support | Oct 2026 | n8n Docs — AI Agent node |
Strengths
- ●520 built-in node pages plus 11,890 community npm packages (July 2026)
- ●Three MCP surfaces, including an instance server that lets an agent build workflows
- ●19 chat-model providers including local Ollama, 13 vector stores and 8 memory backends
- ●Runs self-hosted via npm, Docker or Kubernetes, and embeds under an OEM licence
- ●Source-available and self-hostable
- ●Code nodes plus 2,366 listed integrations, a count that includes core, trigger and AI nodes (n8n.io/integrations, read 13 October 2026)
- ●AI Agent node and AI orchestration direction
- ●More than 200,000 GitHub stars as of September 2026
- ●400+ integrations
- ●Active 250K+ community
- ●Visual + code nodes
- ●Docker one-line install
- ●Excellent documentation
- ●Unlimited self-hosted executions on any infrastructure
- ●900+ integrations with code nodes for custom logic
- ●Active community with more than 200,000 GitHub stars as of September 2026
- ●Self-hosted Community Edition is free; SSO, Git version control and environments start on the paid, self-hosted-only Business plan (September 2026)
- ●Self-hosted with Docker/Kubernetes
- ●JavaScript and Python code nodes
- ●Clean JSON workflow export for Git
- ●More than 200,000 GitHub stars as of September 2026 and an active community
- ●Free self-hosted (unlimited)
- ●Code nodes for custom logic
- ●Active open-source community
- ●Self-hosted on Docker or npm under fair-code license
- ●JavaScript code nodes for custom logic without leaving the canvas
- ●Over 500 integrations, with regular community contributions (n8n.io, read 29 September 2026)
- ●Cloud plan from €20/month billed annually for teams that prefer managed hosting (n8n.io/pricing, read 29 September 2026)
- ●Unlimited free self-hosting
- ●Cloud plans cheaper than Zapier at every tier
- ●Full workflow JSON export
- ●400+ integrations with no premium connector surcharge
- ●Visual ETL pipeline builder
- ●400+ data connectors
- ●Self-hostable for data privacy
- ●Active community with templates
- ●Self-hosted AI pipeline control
- ●LangChain integration
- ●AI agent nodes
- ●500+ integrations with code-in-node flexibility (n8n.io, read 29 Sep 2026)
- ●400+ built-in integrations with an editable node-based visual builder
- ●Code nodes support JavaScript and Python for custom logic within workflows
- ●Queue mode with Redis enables horizontal worker scaling for high throughput
- ●500+ integrations (n8n.io, read 29 Sep 2026)
- ●Self-hosted with no per-ticket or per-agent costs
- ●Full code access (JavaScript/Python) for custom ticket processing logic
- ●900+ integrations including helpdesk and CRM platforms
- ●Cloud Starter is 20 euros a month billed annually for 2,500 executions, the cheapest published meter in this ranking (n8n.io/pricing, 14 September 2026)
- ●The self-hosted Community Edition carries no licence fee, so the gate can run with no vendor invoice at all
- ●The human-in-the-loop step reaches an approver across nine notification channels, so the decision arrives where the approver already works (docs.n8n.io, read 21 September 2026)
- ●The meter is the execution rather than the decision. Adding a second approval to a workflow does not add a second charge
- ●Visual node canvas approachable for non-developers, with optional JavaScript steps
- ●More than 200,000 GitHub stars as of September 2026 and a large library of community node integrations
- ●Self-hostable core lets teams keep all data and credentials inside their network
- ●Built-in queue mode scales execution horizontally for higher throughput
- ●Self-hosted with zero per-order or per-execution costs
- ●Full code access for custom inventory sync and order routing logic
- ●Data stays on-premises for PCI and privacy compliance
- ●Visual workflow builder accessible to analysts and engineers alike
- ●Self-hosted option with no per-execution costs
- ●JavaScript and Python code nodes for custom transformations
- ●900+ integrations covering most SaaS data sources
- ●520 built-in nodes plus an ecosystem of 11,890 community npm packages (July 2026)
- ●Queue mode with Redis and separate workers scales horizontally
- ●Wait nodes persist to the database and resume on webhook or form submission
- ●206,262 GitHub stars as of 29 September 2026: the most-starred repository in this ranking
- ●Self-hosted deployment keeps all PHI inside the organisation's network
- ●Source-available code is auditable by the hospital security team
- ●206,262 GitHub stars as of 29 September 2026 (api.github.com/repos/n8n-io/n8n) and an active community for healthcare use cases
- ●Custom HTTP and database nodes let IT teams integrate proprietary EHRs
- ●Self-hostable (free)
- ●Code-in-node capability
- ●Active community
- ●Self-hosted deployment keeps all transaction data inside the fintech's VPC
- ●Source-available for security-team review and on-prem compliance audits
- ●Custom HTTP and database nodes integrate with internal and core-banking services
- ●Business plan adds SSO, environments and Git version control; Enterprise adds log streaming (n8n.io/pricing, read 29 September 2026)
- ●MCP Server Trigger and MCP Client Tool nodes are built in rather than add-ons
- ●Self-hosting keeps agent credentials and execution on infrastructure the team controls
- ●Roughly 520 built-in nodes cover the common business SaaS surface
- ●Cloud plans start well below the enterprise platforms in this ranking
Limitations
- ●Full code control requires self-hosting: NODE_FUNCTION_ALLOW_EXTERNAL is an environment variable Cloud users cannot set
- ●Unverified community nodes are unavailable on n8n Cloud
- ●On Cloud the Python option imports no libraries at all, standard library included
- ●On n8n Cloud, custom roles, SSO, Git source control and environments require Enterprise; on self-hosted n8n, SSO, Git source control and environments start on the Business plan (September 2026)
- ●n8n Cloud runs in an Azure EU region with no US option
- ●Self-hosting adds operational overhead
- ●Steeper learning curve than linear builders
- ●Sustainable Use License (not OSI-approved)
- ●Cloud version required for some enterprise features
- ●Steeper learning curve for non-technical users
- ●Self-hosting requires Docker and basic server administration
- ●Fair-code license restricts commercial redistribution
- ●Cloud free tier is limited to 5 active workflows
- ●Learning curve is steeper than simpler alternatives
- ●Fair-code license has commercial limitations
- ●TypeScript support is indirect via code nodes
- ●No native multi-language support beyond JS/Python
- ●Requires VPS for self-hosting
- ●Steeper learning curve than no-code tools
- ●Cloud pricing less competitive
- ●Self-hosted operators handle their own backups, scaling, and updates
- ●License is fair-code rather than fully OSI-approved open source
- ●Self-hosting requires technical capacity
- ●Smaller integration library than Zapier
- ●Learning curve steeper than simpler tools
- ●Not purpose-built for ETL
- ●Large dataset handling limitations
- ●Memory constraints on big transforms
- ●AI features require technical setup
- ●No managed AI agent builder
- ●Fair-code (Sustainable Use) license restricts commercial redistribution — not OSI-approved open source
- ●Workflow persistence defaults to SQLite, which creates bottlenecks above ~5,000 executions/day without PostgreSQL
- ●Visual-first editor is less code-reviewable than SDK-based engines; diffs in JSON are harder to audit
- ●AI Workflow Builder is labelled Preview and is credit-metered (docs.n8n.io, read 29 Sep 2026)
- ●Self-hosting adds ops overhead
- ●Requires technical team to set up and maintain workflows
- ●No native helpdesk features — purely a workflow automation layer
- ●Leave the approver list empty and anyone who can see the message can approve, so the gate is only as strong as the channel it was posted to
- ●By default the node's output records only "approved" and a timestamp, which shows that a decision happened but not who made it
- ●The tool-review decision is binary, with no send-back-for-revision and no named rejection branch, which is why decision routes score 4 out of 10
- ●Sustainable Use Licence is more restrictive than pure Apache 2.0 for SaaS resale
- ●Durable execution semantics are visual-flow oriented, not history-replay like Temporal
- ●Production governance (versioning, RBAC, secrets) is less mature than enterprise iPaaS
- ●Requires technical team for setup and maintenance
- ●No pre-built ecommerce templates — workflows built from scratch
- ●Not designed for high-volume batch data processing
- ●No native data warehouse connectors or schema management
- ●Lacks DAG dependency management for complex pipelines
- ●Single-node execution limits throughput for large datasets
- ●No parallel execution: nodes run one at a time and branches execute depth-first
- ●No durable execution; a crashed run is marked crashed and a retry replays from the start
- ●Retries are capped at 5 attempts with a fixed 5,000 ms delay and no backoff
- ●Git source control, environments and log streaming are Enterprise-only
- ●The Sustainable Use License permits internal business and non-commercial use only
- ●No vendor-signed BAA on the self-hosted Community edition — compliance is the operator's responsibility
- ●Requires in-house platform engineering to run, secure, and patch
- ●Fewer prebuilt clinical-system templates than UiPath or Boomi
- ●Fair-code license limitations for resale
- ●UI less polished than commercial tools
- ●Smaller enterprise feature set
- ●No vendor-signed SOC 2 or PCI DSS on the self-hosted edition — compliance belongs to the operator
- ●Needs in-house platform engineering to run, monitor, and patch
- ●Few prebuilt fintech connectors; core-banking and KYC work runs through custom HTTP and database nodes
- ●Agents reach only workflows that have been built and published first
- ●Roughly 520 built-in nodes is a fraction of the catalogues at the top of this list
- ●Self-hosting moves reliability, credential security and rate-limit handling onto the operator
- ●The 2,018-entry directory count uses a different counting basis and overstates application coverage
Based on evaluations in 22 rankings: Most Customizable AI Agent Platforms 2026, Best Workflow Automation Tools 2026, Best Open Source Automation Platforms 2026, Best Free Automation Tools in 2026, Best Automation Tools for Developers and Engineering Teams in 2026, Best Automation Tools for Startups in 2026, Best No-Code Automation Platforms in 2026, Best Automation Tools by Price-Performance Value in 2026, Best ETL & Data Pipeline Tools 2026, Best AI Agent Platforms in 2026, Best Open-Source Workflow Engines for Engineers in 2026, Best AI-Powered Automation Tools in 2026, Best Customer Support Automation Tools in 2026, Best Human-in-the-Loop Automation Platforms 2026, Best Durable Workflow Engines for Production in 2026, Best Automation Tools for Retail & Ecommerce in 2026, Best Automation Tools for Data Teams in 2026, Best Process Orchestration Platforms 2026, Best Automation Tools for Healthcare in 2026, Best iPaaS and Integration Platforms 2026, Best Automation Tools for Fintech and Financial Services in 2026, Best Automation Platforms for AI Orchestration 2026
Pricing Plans
Community
Self-hosted, open fair-code
- ✓Unlimited workflows + executions
- ✓All integration nodes
- ✓JavaScript + Python in nodes
- !Self-managed infrastructure
Starter (Cloud)
($20.00/mo billed annually)
$24/mo monthly, ~$20/mo billed annually
- ✓Managed cloud hosting
- ✓5 active workflows
- !2,500 executions/month
Pro (Cloud)
($50.00/mo billed annually)
$60/mo monthly, ~$50/mo billed annually
- ✓Unlimited active workflows
- ✓Execution debugging, global variables
- !10,000 executions/month
Business
($667.00/mo billed annually)
$800/mo monthly, ~$667/mo billed annually; cloud or self-hosted
- ✓30 concurrent executions
- ✓6 shared projects, 30-day insights
- ✓Self-hosted option
- !40,000 executions/month
Enterprise
Contact sales; hosted or self-hosted
- ✓SSO/SAML, admin panel
- ✓200+ concurrent, 365-day insights
- ✓Dedicated support
About n8n
n8n is a workflow-automation and AI-orchestration platform that connects applications and services through a visual, node-based editor while allowing arbitrary code inside any step. It was founded in 2019 by Jan Oberhauser and is based in Berlin, Germany. Its defining characteristic is the fair-code model: the source is publicly available and free to self-host, which sets it apart from the fully proprietary, cloud-only design of Zapier and Make.
A workflow in n8n is a graph of nodes, where each node is either a trigger, an app integration, a logic operation, or a code step. Code nodes run JavaScript or Python and can import libraries, so a builder is never boxed in by the visual layer when a transformation needs real logic. The platform lists more than 1,830 integrations as of 2026, and because the node framework is open, integrations that a vendor has not built can be added by the community or written in-house. Billing on n8n Cloud is counted per workflow execution: one execution is a single run of an entire workflow regardless of how many nodes it contains or how much data passes through, which is a materially different cost model from per-task or per-operation competitors.
From 2024 onward n8n repositioned from "workflow automation" toward "AI orchestration." The flagship addition is the AI Agent node, which bundles a chat model, a system prompt, memory, and callable tools into a single step, and an AI Workflow Builder that generates flows from natural-language prompts using a per-tier credit allotment. This direction is the reason n8n is now frequently evaluated alongside dedicated AI-agent platforms rather than only against integration tools.
n8n ships in two main forms. The Community edition is free to self-host, distributed on GitHub under the Sustainable Use License (source-available, free for internal and personal use, with resale as a hosted service requiring authorization). n8n Cloud is the managed offering: as of June 2026 Cloud Starter is €20 per month for 2,500 executions, Cloud Pro is €50 per month for 10,000 executions, Business is €667 per month, and Enterprise is custom-priced. The permanent free Cloud plan was discontinued in favour of a time-limited trial, so the genuinely free path is self-hosting the Community edition.
Diagram source (Mermaid)
flowchart TD
A[Choosing an n8n deployment] --> B{Data must stay in your infrastructure?}
B -- Yes --> C[Community edition: self-host, free, Sustainable Use License]
B -- No --> D{Need SSO, Git sync, shared projects?}
D -- Yes --> E[Cloud Business EUR667/mo or Enterprise]
D -- No --> F{Monthly executions}
F -- Up to ~2,500 --> G[Cloud Starter EUR20/mo]
F -- Up to ~10,000 --> H[Cloud Pro EUR50/mo]
n8n competes with Zapier, Make, Pipedream, and Power Automate, and increasingly with AI-agent platforms. Its commercial momentum is unusual for the category: the company raised a $180M Series C in October 2025, led by Accel with participation from NVIDIA's NVentures, at a post-money valuation reported around $2.5 billion, on the back of steep 2025 growth in users and revenue. The trade-off buyers weigh is operational: self-hosting removes per-execution cost but adds the work of running, upgrading, and monitoring the system, while n8n Cloud removes that work in exchange for execution-based billing.
Editor's Note: Across our n8n self-hosting deployments at ShadowGen in 2024-26, the recurring win is cost and the recurring trap is data retention. One mid-market SaaS client ran roughly 2,300 executions per day on a single 2 vCPU / 4 GB VPS for about $12 per month, replacing a Zapier plan several multiples of that. The caveat that bit us once: the execution-log table grew past 40 GB before anyone set EXECUTIONS_DATA_MAX_AGE and pruning, which quietly degraded the Postgres backing store. Set retention on day one, not after the first slowdown. — Rafal Fila, ShadowGen
Integrations (6)
Other Workflow Automation Tools
Activepieces
No-code workflow automation with self-hosting and AI-powered features
Workflow AutomationAutomatisch
Open-source Zapier alternative
Workflow AutomationBardeen
AI-powered browser automation via Chrome extension
Workflow AutomationCalendly
Scheduling automation platform for booking meetings without email back-and-forth, with CRM integrations and routing forms for lead qualification.
Workflow AutomationSee How It Ranks
Best Human-in-the-Loop Automation Platforms 2026
Eight platforms are ranked here on a single mechanism: how a person enters an automated path, and what the system can prove afterward about the decision they made. Moxo scores 7.7 and ranks first; Camunda 8 scores 7.2, Microsoft Power Automate 7.1, Pega 6.9, n8n 6.7, Zapier 5.9…
Best Automation Platforms for AI Orchestration 2026
This ranking answers one question: how many real business applications can an AI agent act on out of the box? It evaluates nine platforms as of August 2026 on the reach they give an agent, not on the workflow logic they can express. That boundary is deliberate, because two…
Questions About n8n
Can you build n8n workflows with Claude Code?
Yes. As of 26 August 2026 n8n ships a built-in instance-level MCP server that Claude Code connects to over streamable HTTP, and n8n's documentation lists Claude Code by name among its supported CLI clients alongside Codex and Gemini CLI. Connect it with `claude mcp add…
How much does it cost to give an AI agent access to your automation platform?
The cost turns entirely on the platform's billing unit, and as of 26 August 2026 only Zapier publishes an explicit per-call rate. Zapier's documentation states that "Each successful tool call through your MCP server consumes two tasks", while "Tool calls that fail do not consume…
What should teams do now that Relay.app is shutting down?
Relay.app announced on 16 July 2026 that it is shutting down. Free accounts and all their data are permanently deleted after 15 August 2026 at 23:59 PT, and paid accounts after 14 September 2026 at 23:59 PT, with paying customers keeping full access at no charge until that date…
What is Keystroke?
Keystroke is a code-first workflow automation and AI agent platform, launched July 13, 2026 by Y Combinator-backed Sprint Labs, that positions itself as an n8n alternative built for AI coding agents. Workflows are written as typed TypeScript in the user's own repository, usually…
Learn More
Approving an AI Agent's Action Mid-Workflow: Where the Human Gate Goes
This guide sets out four places a human gate can sit when an AI agent acts in a workflow (before the action, after the draft, on exceptions only, or by sampling afterwards) and how Moxo implements each, from its MCP docs and OpenAPI copy read 13 October 2026. It also covers…
Moxo vs n8n in 2026: Who Runs the Engine and Who Keeps the Record
n8n wins the engine room: Community Edition runs on a self-hosted server at no licence cost, and its code is open to read and change for internal use. Moxo wins the approval itself: its API refuses a step completion from an organisation key, a stalled step has named moves, and…
AI Coding Agents and Automation Platforms: Three Integration Models
This guide classifies the ways an AI coding agent can connect to a business process automation platform into three integration models: reference servers that teach an agent a platform's schema, control-plane servers that let it read and edit workflows on a live instance, and…