Solanasis Docs

solanasis-n8n-openclaw-langgraph-comparison-playbook-handoff-2026-03-18

24 min read

Solanasis Research Memo: n8n vs OpenClaw vs LangGraph

Verified comparison, cost model, implementation playbook, and AI handoff document
Date: 2026-03-18

Executive Summary

This document extracts, organizes, verifies, and improves the key conclusions from the discussion comparing n8n, OpenClaw, and LangGraph / LangSmith for Solanasis use cases.

Top-line conclusions

  1. [Verified] n8n is a workflow automation platform, not primarily an agent runtime.
    n8n describes itself as a fair-code workflow automation tool that combines AI capabilities with business process automation. It is strongest when the work is trigger-based, integration-heavy, and operationally repeatable.[S1]

  2. [Verified] LangGraph is a low-level orchestration framework and runtime for long-running, stateful agents.
    LangChain’s documentation explicitly positions LangGraph for durable execution, human-in-the-loop control, memory, and long-running stateful workflows/agents.[S2][S3][S4]

  3. [Verified] OpenClaw is a local-first personal assistant/control plane, not a multi-tenant business automation platform by default.
    OpenClaw’s docs position it as a personal AI assistant running on your own devices and reachable through real chat surfaces such as WhatsApp, Telegram, Slack, Discord, Signal, and iMessage. Its security model is explicitly “personal assistant,” not a general multi-tenant authorization boundary.[S5][S6][S7]

  4. [Conclusion; strongly grounded] For most early Solanasis client work, n8n is the better first operating layer than LangGraph.
    Reason: the client problems described for Solanasis are primarily around operational workflows, integrations, approvals, evidence collection, CRM/process automation, disaster recovery verification, and bounded AI. Those are closer to workflow automation than to stateful agent engineering. This is not because LangGraph is weak; it is because it is designed for a different center of gravity.[S1][S2][S3][S8]

  5. [Conclusion; strongly grounded] LangGraph becomes the better foundation when the agent itself is the product or when durable state is core to the workflow.
    LangChain explicitly says to use LangGraph when you need durable execution, low-level control, and production-ready infrastructure for agent deployment.[S4]

  6. [Verified] n8n’s official Anthropic integration is API-key based, not Claude subscription-auth based.
    n8n’s Anthropic credentials docs instruct users to create an API key in the Anthropic Console.[S9]

  7. [Verified] A Claude Max/Pro subscription proxy path exists, but it is community-based and policy-sensitive.
    OpenClaw documents a community tool, claude-max-api-proxy, that exposes a Claude Max/Pro subscription as an OpenAI-compatible API endpoint, but explicitly warns that Anthropic has blocked some subscription usage outside Claude Code in the past and that users should verify Anthropic’s current terms before relying on it.[S10]

  8. [Verified] n8n can be safely exposed as a constrained tool layer via MCP, but MCP-triggered workflows have important limitations.
    n8n requires workflows to be explicitly published and enabled for MCP access, supports OAuth2 or access tokens, can revoke client access, and can disable MCP entirely. Limitations include a 5-minute timeout, text-only inputs, no binary input, and no human-in-the-loop/multi-step form workflows through MCP execution.[S11]

  9. [Conclusion; strongly grounded] “More production-ready” is context-specific.
    LangGraph is more production-ready for stateful agent deployment.
    n8n is more production-ready for integration-centric SMB operations automation.
    Treating those as the same category is the core mistake to avoid.[S1][S2][S4]

Purpose of This Document

This artifact is intended to serve as:

  • a guide for Solanasis internal strategy,
  • a playbook for architecture selection,
  • a briefing memo for business and technical decision-making,
  • and a handoff document for another AI so the work can continue without the original chat.

This document does not assume that every claim in the original discussion was correct. It separates:

  • Verified
  • User-stated
  • Assistant-stated but unverified
  • Tentative / speculative

Discussion Context

User goals and constraints

  • [User-stated] Solanasis is evaluating n8n as both:
    • a relevant platform for Solanasis client services, and
    • a “safer OpenClaw alternative” for controlled automation.
  • [User-stated] The user also wants to understand whether n8n can be run using a “local clawed code” / Claude subscription-style setup instead of pure API billing.
  • [User-stated] The user then asked how n8n compares to LangGraph, including cost, production readiness, and why one would not simply go straight to LangGraph.

Implied Solanasis decision frame

  • [Conclusion; strongly grounded] The practical decision is not “which tool is better overall?”
    It is:

    1. Which tool best fits Solanasis’s client delivery model?
    2. Which tool is safer and easier to govern for SMB/nonprofit operations?
    3. Which tool is appropriate for internal founder-assistant workflows versus client-facing production systems?
    4. Which tool keeps the door open for future, more agentic systems without forcing premature complexity?

Evidence Legend

LabelMeaning
VerifiedConfirmed by primary or official source linked in this document
User-statedCame from the user’s prompt or business context; not independently verified here
Assistant-stated but unverifiedPreviously stated in discussion, but not fully verified in this artifact
Tentative / speculativeReasoned inference, recommendation, or hypothesis rather than a confirmed fact

Key Facts and Verified Findings

A. What n8n is

PointStatusNotesSources
n8n is a fair-code workflow automation tool that combines AI capabilities with business process automation.VerifiedCore official description.[S1]
n8n helps connect apps/APIs and manipulate data with little or no code.VerifiedCore product definition.[S1]
n8n supports self-hosting and cloud hosting.VerifiedOfficial docs cover both.[S12]
n8n recommends self-hosting for expert users and warns of risks such as data loss, security issues, and downtime if done poorly.VerifiedImportant operational caveat for Solanasis.[S12]
n8n’s self-hosted Community Edition is free.VerifiedImportant cost/go-to-market fact.[S12]
n8n is fair-code, not OSI open source.VerifiedImportant for white-label/resale strategy.[S13]
n8n’s license allows consulting/support/workflow-building services, but restricts white-labeling/hosting n8n as a paid product where value derives substantially from n8n itself.VerifiedImportant commercial constraint.[S13]

B. What LangGraph is

PointStatusNotesSources
LangGraph is a low-level orchestration framework and runtime for long-running, stateful agents.VerifiedOfficial LangChain positioning.[S4]
LangGraph is designed for durable execution, human-in-the-loop, memory, and long-running workflows.VerifiedExplicit in official docs.[S2][S3]
LangChain says to use LangGraph when you need fine-grained control, durable execution, complex deterministic+agentic workflows, and production-ready infrastructure for agent deployment.VerifiedThis is the best official “when to use it” summary.[S4]
LangGraph itself is MIT-licensed open source.VerifiedImportant licensing distinction from n8n.[S14]
Agent Server deployments involve one or more graphs, a persistence database, and a task queue.VerifiedImportant ops complexity note.[S15]
Standalone LangGraph agent servers can be self-hosted without the LangSmith control plane, but you must manage your own infrastructure/configuration.VerifiedImportant nuance: framework is free, production ops are still real.[S16]

C. What OpenClaw is

PointStatusNotesSources
OpenClaw is positioned as a personal AI assistant you run on your own devices.VerifiedOfficial wording.[S5]
OpenClaw works across real messaging surfaces such as WhatsApp, Telegram, Slack, Discord, Signal, iMessage, and others.VerifiedOfficial docs/GitHub repo.[S5][S7][S17]
OpenClaw emphasizes a local-first control plane, stateful sessions, memory, and multi-agent routing.VerifiedOfficial positioning.[S5][S7]
OpenClaw’s security model is “personal assistant,” not a general multi-tenant bus.VerifiedCrucial safety caveat.[S6]
If several people can message one tool-enabled OpenClaw agent, they can steer the same permission set.VerifiedThis is a major operational warning.[S6]
If users may be adversarial to one another, OpenClaw recommends separate gateways or separate OS users/hosts per trust boundary.VerifiedImportant for client deployment safety.[S6]

D. n8n MCP and AI workflow facts

PointStatusNotesSources
n8n has an MCP server capability.VerifiedOfficial docs.[S11]
MCP access can be disabled entirely via N8N_DISABLED_MODULES=mcp.VerifiedGood hardening option.[S11]
n8n supports OAuth2 or access-token auth for MCP clients.VerifiedImportant control-plane detail.[S11]
No workflows are exposed to MCP by default; eligible workflows must be explicitly enabled.VerifiedImportant safety default.[S11]
Eligible workflows must be published and use a supported trigger such as Webhook, Schedule, Chat, or Form.VerifiedImportant exposure rule.[S11]
MCP-triggered workflows have a 5-minute timeout.VerifiedStrong practical limitation.[S11]
MCP-triggered workflows do not support multi-step forms, HITL interactions, or binary input.VerifiedImportant comparison point vs LangGraph interrupts.[S11]
n8n supports human review before AI tool execution inside n8n AI workflows.VerifiedStrong governance feature.[S18]
n8n can route approval requests through channels such as Slack, Telegram, or n8n Chat.VerifiedUseful for governance design.[S18]
n8n warns that community nodes have full machine access and access to workflow data.VerifiedImportant security risk.[S19]
n8n lets admins block specific nodes (e.g., Execute Command, Read/Write Files) with NODES_EXCLUDE.VerifiedStrong hardening feature for client environments.[S20]

E. Claude / Anthropic subscription vs API facts

PointStatusNotesSources
n8n’s official Anthropic credential flow uses API keys from the Anthropic Console.VerifiedFirst-party integration path.[S9]
OpenClaw supports Anthropic API keys or claude setup-token authentication.VerifiedRelevant for OpenClaw, not official n8n.[S21][S22]
OpenClaw documents a community tool (claude-max-api-proxy) that exposes Claude Max/Pro as an OpenAI-compatible endpoint.VerifiedImportant but policy-sensitive.[S10]
OpenClaw warns this proxy path is “technical compatibility only” and says users should verify Anthropic’s current terms before relying on it.VerifiedKey caution.[S10]
n8n’s generic HTTP Request credentials support common auth methods including bearer auth, OAuth2, header auth, etc.VerifiedThis is what makes proxy integration technically possible.[S23]

F. Cost facts

PointStatusNotesSources
n8n Starter: €20/month billed annually, 2.5K workflow executions with unlimited steps.VerifiedHosted by n8n.[S24]
n8n Pro: €50/month billed annually, custom number of workflow executions.VerifiedHosted by n8n.[S24]
n8n Business: €667/month billed annually, 40K workflow executions with unlimited steps, self-hosted.VerifiedIncludes SSO, environments, Git-based version control, scaling options.[S24]
n8n Community Edition is free self-hosted.VerifiedOfficial.[S12]
LangSmith Developer: $0/seat/month, up to 5k base traces/month.VerifiedLangGraph platform-adjacent managed cost.[S25]
LangSmith Plus: $39/seat/month, up to 10k base traces/month, 1 dev-sized deployment included.VerifiedOfficial pricing.[S25]
Additional LangSmith deployment runs are $0.005/run.VerifiedOfficial pricing.[S25]
LangSmith deployment uptime is 0.0036/min for production.VerifiedOfficial pricing.[S25]
Base traces cost 5.00 per 1k.VerifiedOfficial pricing FAQ language.[S26]
LangSmith bills LLM usage separately through the model provider.VerifiedImportant total-cost caveat.[S25]
Arithmetic example: one always-on production deployment at 155.52 over a 30-day month.Verified calculation based on official pricing43,200 min × $0.0036.Derived from [S25]
Arithmetic example: one always-on development deployment at 30.24 over a 30-day month.Verified calculation based on official pricing43,200 min × $0.0007.Derived from [S25]

Major Decisions and Conclusions

1. n8n and LangGraph are not substitutes in the strict sense

  • [Verified] n8n is built around workflow automation, integrations, and business process automation.[S1]
  • [Verified] LangGraph is built around durable, stateful agent orchestration.[S2][S4]
  • [Conclusion; strongly grounded] Comparing them as if they are direct equivalents creates confusion. They overlap in some AI workflow scenarios, but their center of gravity is different.

2. Solanasis should not default to LangGraph just because it sounds more “production-ready”

  • [Verified] LangChain says LangGraph is production-ready for agent deployment.[S4]
  • [Conclusion; strongly grounded] That does not automatically make it the best first platform for Solanasis’s immediate client services, which are mostly operational workflow problems, not stateful agent-product problems.
  • [Conclusion; strongly grounded] For Solanasis, “production-ready” should mean:
    • safe enough for client operations,
    • auditable enough for compliance-sensitive clients,
    • cheap enough to deploy repeatedly,
    • understandable enough to hand off,
    • and constrained enough to avoid agent sprawl.

3. n8n is usually the better first choice for Solanasis service delivery

  • [Tentative / speculative, but strongly grounded] n8n is the better near-term default for:
    • security assessment intake and evidence routing,
    • disaster recovery verification reminders and reporting,
    • CRM and back-office integration,
    • client onboarding flows,
    • bounded AI approvals,
    • nonprofit operational automation,
    • internal AI-assisted process automation.

Why:

  • [Verified] n8n has first-class workflow triggers, integrations, AI workflow support, approvals, and MCP exposure controls.[S1][S11][S18]
  • [Verified] n8n can be self-hosted and hardened for privacy/security, and can block risky nodes.[S12][S20]
  • [Verified] n8n’s Community Edition can be used free in self-hosted mode, which matters for experimentation and margin.[S12]

4. LangGraph is the better choice once Solanasis is building true agent systems

  • [Tentative / speculative, but strongly grounded] Use LangGraph when one or more of these are true:

    • the agent is the core product,
    • workflow state must persist across long periods,
    • interruption/resume is fundamental,
    • multi-agent orchestration is central,
    • custom control over graph/state/checkpointing is required,
    • a code-first team is ready to own an agent runtime.

  • [Verified] These are the exact sorts of needs LangChain documents for LangGraph.[S2][S4]

5. OpenClaw is not the right general-purpose multi-tenant client automation backbone

  • [Verified] OpenClaw explicitly warns that if several people can message one tool-enabled agent, they can steer the same permission set.[S6]
  • [Verified] OpenClaw’s security model is personal-assistant-centric, not a per-user authorization system.[S6]
  • [Conclusion; strongly grounded] Therefore OpenClaw is best framed as:
    • an internal founder/copilot/personal-assistant platform,
    • or a carefully bounded single-trust-boundary company assistant,
    • not the default backend for multi-tenant client operations.

Reasoning, Tradeoffs, and Why It Matters

1. Workflow automation vs agent orchestration

n8n’s strength

  • [Verified] Strong at event-driven workflows, integrations, business process automation, API connectivity, and bounded AI.[S1][S23]
  • [Why it matters] Most SMBs/nonprofits are not asking for “an autonomous agent runtime.” They are asking for:
    • fewer dropped tasks,
    • fewer manual handoffs,
    • fewer spreadsheet-to-email-to-CRM messes,
    • and safer AI-assisted process automation.

LangGraph’s strength

  • [Verified] Strong at durable state, interrupts, memory, and low-level orchestration.[S2][S3][S4]
  • [Why it matters] If Solanasis later productizes a persistent AI operator, analyst, or case manager, LangGraph becomes much more compelling.

OpenClaw’s strength

  • [Verified] Strong at chat-surface presence, multi-channel reachability, local-first control, multi-agent routing, and personal-assistant ergonomics.[S5][S7]
  • [Why it matters] Great for founder workflows and internal assistant patterns. Much less ideal as the canonical automation substrate for multiple client teams with varying trust boundaries.

2. Safety and governance

Why n8n often feels “safer”

  • [Verified] n8n can require explicit workflow publishing/MCP enablement.[S11]
  • [Verified] n8n can require human approval before AI tool execution.[S18]
  • [Verified] n8n can exclude risky nodes globally.[S20]
  • [Verified] n8n can disable MCP entirely.[S11]
  • [Conclusion; strongly grounded] This creates a more controllable blast radius than a generally reachable personal assistant gateway.

Why OpenClaw can feel riskier in shared contexts

  • [Verified] Shared tool-enabled agent = shared permission set.[S6]
  • [Verified] OpenClaw recommends separate gateways per trust boundary when adversarial-user isolation matters.[S6]
  • [Conclusion; strongly grounded] That is workable, but it is more operationally delicate for client environments.

Why LangGraph can be safer than OpenClaw for app-grade agent systems

  • [Verified] LangGraph has durable execution and explicit interrupts for HITL.[S2][S3]
  • [Tentative / speculative] When built correctly, that enables very strong governance patterns.
    But it still requires an engineering team to implement those controls correctly; they are not “business workflow defaults” in the way n8n’s UI-driven approval patterns are.

3. Cost model tradeoffs

n8n

  • [Verified] Low entry cost, including free self-hosting.[S12][S24]
  • [Verified] Pricing is primarily based on workflow executions rather than traces/deployment-runtime abstractions.[S24]
  • [Conclusion; strongly grounded] Easier for simple SMB automation economics.

LangGraph / LangSmith

  • [Verified] Framework is MIT/open source.[S14]
  • [Verified] Managed deployment/observability costs come through LangSmith seats, traces, deployment runs, uptime, and model-provider costs.[S25][S26]
  • [Conclusion; strongly grounded] The framework may be “free,” but production ownership is not.

OpenClaw + Claude subscription proxy path

  • [Verified] Community proxy exists.[S10]
  • [Verified] OpenClaw warns to verify policy/terms before relying on it.[S10]
  • [Conclusion; strongly grounded] This can reduce cost for internal experimentation, but it is not the cleanest contractual or technical path for client production systems.

Recommended Playbook / Process

Solanasis architecture selection playbook

Step 1: Classify the client problem

Question A: Is this mostly a repeatable business process / integration problem?
Examples:

  • onboarding workflow
  • CRM sync
  • follow-up reminders
  • evidence intake
  • report routing
  • DR checklist orchestration

Default answer: Start with n8n.

Question B: Is this a persistent agent problem?
Examples:

  • an assistant that must pause, remember, and resume over days/weeks
  • a complex analyst/caseworker-style agent
  • a multi-agent decision flow
  • custom stateful orchestration with a lot of code-level control

Default answer: Consider LangGraph first.

Question C: Is this mostly a founder/internal personal-assistant or chat-surface automation problem?
Examples:

  • “message my assistant in Slack/Telegram/WhatsApp”
  • “do research and follow up from my phone”
  • “route across my channels and tools”
  • “keep my own sessions and workspace local”

Default answer: Consider OpenClaw, but keep it out of shared multi-user client automation unless carefully isolated.

Solanasis default stack recommendation

[Tentative / speculative, strongly grounded]

  1. n8n as the automation backbone
  2. LLM provider via API for production clarity
  3. Human approval gates for risky actions
  4. Client systems of record (CRM, docs, ticketing, database) outside the AI layer
  5. Optional Claude Desktop / Claude Code / other chat client calling narrow workflows over MCP

This gives Solanasis:

  • bounded automations,
  • easier repeatability,
  • lower deployment complexity,
  • clearer auditability,
  • and a safer story for SMB/nonprofit clients.

Safe reference patterns

Pattern A: Safer client workflow automation

Frontend/chat/request → n8n workflow → approval gate → client system/API

Use for:

  • CRM updates
  • client onboarding
  • case routing
  • document generation/delivery
  • backup verification reminders
  • compliance follow-ups

Why this works

  • [Verified] n8n supports workflow gating, approvals, and selective MCP exposure.[S11][S18]

Pattern B: Internal founder copilot + constrained rails

OpenClaw → n8n workflow for narrow actions → human approval for sensitive steps

Use for:

  • Dmitri/internal ops assistant flows
  • quick founder commands
  • chat-native access to safe automations
  • personal/internal experimentation

Why this works

  • Preserves OpenClaw’s convenience while keeping destructive or business-critical logic in narrower workflows.

Pattern C: Advanced agent product pattern

Frontend/app → LangGraph agent runtime → n8n or direct APIs for deterministic subflows → LangSmith or self-hosted observability/deployment

Use for:

  • persistent AI case manager
  • multi-agent orchestration
  • long-running client-facing AI operator
  • systems needing pause/resume/memory as core product behavior

Suggested Solanasis starter use cases for n8n

1. Security assessment intake and evidence routing

  • intake form
  • auto-create project record
  • request evidence from client
  • classify evidence gaps
  • schedule reminders
  • generate exec-summary draft
  • approval before sending final summary

2. Disaster recovery verification workflow

  • recurring scheduled trigger
  • pull backup/restore status via API where available
  • prompt staff/client for restore evidence
  • flag overdue tests
  • generate monthly verification summary

3. CRM and sales ops workflow

  • inbound lead capture
  • enrichment/qualification
  • route to CRM
  • generate follow-up task/email draft
  • notify internal owner
  • produce pipeline hygiene reminders

4. Nonprofit / special-assets ops automation

  • application/questionnaire intake
  • document collection
  • exception handling
  • compliance reminder routing
  • AI summarization of client status for internal review

Pro Tips and Implementation Notes

n8n setup tips for Solanasis

  1. [Verified + recommendation] Treat n8n like production infrastructure, not a hobby bot.
    Self-hosting requires real technical competence, and n8n says so directly.[S12]

  2. [Verified + recommendation] Use SSL, SSO where appropriate, and 2FA.
    n8n’s security docs explicitly recommend these controls.[S27]

  3. [Verified + recommendation] Block risky nodes in client environments.
    Use NODES_EXCLUDE for things like Execute Command or Read/Write Files if a client deployment does not need them.[S20]

  4. [Verified + recommendation] Be highly cautious with community nodes.
    n8n warns they have full machine access and access to workflow data.[S19]

  5. [Recommendation] Default to built-in nodes and HTTP Request before community nodes.
    This is a practical governance choice grounded in [S19] and [S23].

  6. [Verified + recommendation] Use human review for destructive, external, or high-value actions.
    n8n’s HITL tool approvals are well-suited for this.[S18]

  7. [Recommendation] Separate dev/stage/prod early for any serious client deployment.
    This is especially important once multiple team members or contractors are touching workflows.

  8. [Recommendation] Use service accounts with minimal permissions for integrations.
    Do not let one credential span multiple unrelated systems unless there is a very good reason.

  9. [Verified + recommendation] If experimenting with local AI, use n8n’s self-hosted AI starter kit as a proof-of-concept path, but know the Ollama node lacks tools support for the AI Agent node.
    That makes some “fully agentic local replacement” assumptions invalid.[S28][S29]

  10. [Recommendation] Use API-key-based model access in production until there is a clearly supported, contractually clean alternative.

LangGraph setup tips for Solanasis

  1. [Verified + recommendation] Only use LangGraph when durable state/control is actually needed.
    Otherwise you are buying complexity you do not yet need.[S4]

  2. [Verified + recommendation] Design for deterministic replay and idempotency.
    LangGraph’s durable execution docs explicitly require thought around side effects and replay.[S3]

  3. [Verified + recommendation] Use interrupts for critical approvals or review/edit checkpoints.
    This is one of LangGraph’s best differentiators versus workflow tools.[S2]

  4. [Verified + recommendation] Budget for deployment/runtime/observability, not just framework cost.
    LangGraph itself is MIT; production operations are not free.[S14][S25][S26]

  5. [Verified + recommendation] Expect more engineering ownership.
    Standalone/self-hosted deployments require managing your own infrastructure, and agent server deployments include persistence and task queues.[S15][S16]

OpenClaw setup tips for Solanasis

  1. [Verified + recommendation] Treat OpenClaw as internal or single-trust-boundary by default.
    Do not assume per-user isolation from session/memory separation alone.[S6]

  2. [Verified + recommendation] Use separate gateways or dedicated runtimes per trust boundary if multiple actors can steer tools.
    This is straight from OpenClaw’s security model.[S6]

  3. [Recommendation] Use OpenClaw for founder productivity and chat-native coordination, not as the default multi-client ops automation backbone.

  4. [Verified + recommendation] If testing Claude Max/Pro proxying, keep it in experimental/internal lanes until policy and reliability questions are resolved.[S10]

Core official resources

Risks, Caveats, and Red Flags

n8n risks

  • [Verified] Self-hosting is not trivial and can create security/downtime risk if done poorly.[S12]
  • [Verified] Community nodes can fully access the host machine and workflow data.[S19]
  • [Verified] MCP-triggered workflows cannot do everything; timeout and interaction limitations matter.[S11]
  • [Verified] Licensing is fair-code, not standard open source; embedding/white-label resale has limits.[S13]

LangGraph risks

  • [Verified] Production deployment requires real engineering maturity and infrastructure ownership unless you adopt paid deployment tooling.[S15][S16]
  • [Verified] Durable execution requires deterministic/idempotent design or you can create replay/side-effect bugs.[S3]
  • [Conclusion; strongly grounded] It is easy to over-engineer SMB ops problems by reaching for LangGraph too early.

OpenClaw risks

  • [Verified] Shared agent permissions can be steered by any allowed sender.[S6]
  • [Verified] It is not designed as a general multi-tenant authorization boundary.[S6]
  • [Verified] Claude Max/Pro proxy path is community-based and policy-sensitive.[S10]

Cross-cutting risks

  • [Tentative / speculative] Tool sprawl and credential sprawl are likely bigger risks than model choice for Solanasis client deployments.
  • [Tentative / speculative] “AI-native” client delivery can become fragile fast if approvals, rollback, and source-of-truth architecture are not designed first.
  • [Tentative / speculative] Wealth-management-adjacent and nonprofit clients may impose stronger expectations around auditability, data location, approval controls, and contractor access than typical SMB automation buyers.

Open Questions / What Still Needs Verification

  1. [Open] Anthropic’s current policy position on subscription-backed proxy usage outside Claude Code should be checked directly against current Anthropic terms before any production use.
  2. [Open] Real-world reliability of n8n MCP with Claude Code, Claude Desktop, and Codex CLI should be validated in a Solanasis sandbox, especially around auth rotation, latency, and workflow UX.
  3. [Open] Solanasis should define its preferred deployment model:
    • Solanasis-managed internal instance
    • client-owned instance
    • hybrid managed service
  4. [Open] Exact margin model for managed n8n delivery should be tested with a few reference workflows and client support assumptions.
  5. [Open] Need a policy for when to escalate from n8n to LangGraph.
  6. [Open] Need a credential and secrets-management standard for client automations.
  7. [Open] Need backup/DR standards for Solanasis-managed automation infrastructure itself.
  8. [Open] Need to verify which target clients (SMBs, nonprofits, special-assets/foundation ops, RIAs, wealth-management-adjacent firms) have the strongest fit for bounded workflow automation vs agentic systems.

Suggested Next Steps

Immediate

  1. Build a Solanasis decision matrix:

    • n8n vs LangGraph vs OpenClaw
    • use case
    • trust boundary
    • deployment complexity
    • cost pattern
    • approval requirements

  2. Stand up a small self-hosted n8n sandbox with:

    • SSL
    • 2FA
    • restricted nodes
    • one safe demo workflow
    • one approval-gated AI workflow

  3. Prototype 3 service-grade n8n workflows:

    • DR verification
    • security assessment intake/evidence routing
    • CRM/opportunity follow-up

  4. Create a policy memo for internal use:

    • “No subscription-proxy path in production until terms are verified”
    • “OpenClaw internal-first, not default client backbone”
    • “LangGraph only for durable-state agent cases”

Short-term

  1. Create a reference architecture deck for prospects:

    • human-in-the-loop
    • bounded AI
    • approvals
    • auditability
    • privacy-preserving self-hosting options

  2. Build a pricing model for:

    • setup fee
    • managed automation retainer
    • support/monitoring tier
    • change-request model

  3. Test whether Claude Desktop/Claude Code ↔ n8n MCP is strong enough to be part of the Solanasis internal operator workflow.

Medium-term

  1. Identify one candidate use case where LangGraph would actually create advantage over n8n and prototype it separately, rather than forcing it into all automation work.

Handoff Notes for Another AI

You are inheriting a strategy thread about n8n vs OpenClaw vs LangGraph for Solanasis, a fractional CIO/CISO/COO firm targeting SMBs, nonprofits, and adjacent operational-resilience clients.

What has already been established

  • n8n is the likely best first automation substrate for many Solanasis delivery use cases.
  • LangGraph is likely the right escalation path for true stateful agent systems.
  • OpenClaw is useful, but should be treated primarily as:
    • founder/internal assistant infrastructure, or
    • a tightly bounded single-trust-boundary assistant setup.
  • Claude subscription proxying is not a clean production assumption.

What another AI should do next

  1. Turn the conclusions into a Solanasis-specific packaging/pricing framework.
  2. Map n8n-first offerings into concrete client service packages.
  3. Produce a reference architecture and security checklist for self-hosted n8n.
  4. Identify a single example where LangGraph is clearly justified and design that separately.
  5. Preserve the distinction between:
    • workflow automation
    • agent orchestration
    • personal assistant control planes

Important constraints to preserve

  • Do not flatten all three tools into one category.
  • Do not assume subscription-auth proxying is safe/policy-clean.
  • Do not treat OpenClaw as multi-tenant-safe by default.
  • Do not recommend LangGraph first unless durable state is genuinely central to the problem.
  • Keep Solanasis positioned around safe, bounded, auditable AI-native operations, not “wild autonomous agents.”

Reviewer Notes and Improvements Made

Reviewer mode used: serious self-review pass (no separate reviewer agent was available in this environment).

Improvements made beyond the original discussion

  1. Verified key technical and pricing claims against official documentation.
  2. Reframed the comparison around workflow automation vs agent orchestration vs personal assistant control plane, which is the core conceptual distinction.
  3. Added licensing analysis, including why n8n’s fair-code model matters for resale/white-label strategy.
  4. Added security/governance implications from OpenClaw’s own security model.
  5. Added missing caution around n8n MCP limitations.
  6. Added missing nuance that LangGraph itself is MIT/free, but LangSmith deployment/observability adds real production costs.
  7. Converted the discussion into a Solanasis-usable playbook with clear next steps.
  8. Flagged unresolved areas that require direct validation rather than confident guessing.

Self-review checks performed

  • Checked for unsupported certainty
  • Removed or downgraded broad popularity claims that were not directly verifiable
  • Preserved uncertainty around policy-sensitive proxy use
  • Distinguished verified facts from recommendations and inference
  • Added missing implementation and governance notes that logically follow from the source material

Optional Appendix: Structured Summary (YAML-style)

document:
  title: "Solanasis Research Memo: n8n vs OpenClaw vs LangGraph"
  date: "2026-03-18"
  purpose:
    - guide
    - playbook
    - briefing memo
    - AI handoff artifact
 
core_conclusions:
  - statement: "n8n is the likely best first platform for many Solanasis client automations."
    status: "Tentative / speculative, strongly grounded"
  - statement: "LangGraph is the better fit for durable, stateful agent systems."
    status: "Verified + conclusion"
  - statement: "OpenClaw is best treated as a personal/internal assistant platform unless isolated carefully."
    status: "Verified + conclusion"
  - statement: "Claude subscription proxying is not a clean production default."
    status: "Verified + conclusion"
 
tool_roles:
  n8n:
    type: "workflow automation platform"
    best_for:
      - operational workflows
      - integrations
      - approvals
      - bounded AI
      - SMB/nonprofit automation
    not_best_for:
      - deeply stateful long-running agents
  langgraph:
    type: "stateful agent orchestration framework/runtime"
    best_for:
      - durable execution
      - interrupts
      - memory
      - multi-agent systems
      - custom agent products
    tradeoff:
      - more engineering complexity
  openclaw:
    type: "local-first personal assistant control plane"
    best_for:
      - founder productivity
      - chat-native assistant flows
      - multi-channel assistant access
    risk:
      - not default multi-tenant-safe
 
cost_snapshot:
  n8n:
    community: "free self-hosted"
    starter: "€20/mo annual, 2.5K executions"
    pro: "€50/mo annual, custom executions"
    business: "€667/mo annual, 40K executions self-hosted"
  langsmith:
    developer: "$0/seat, 5k traces"
    plus: "$39/seat, 10k traces, 1 dev deployment"
    extra_runs: "$0.005/run"
    dev_uptime: "$0.0007/min"
    prod_uptime: "$0.0036/min"
 
next_actions:
  - "Build n8n sandbox"
  - "Prototype 3 Solanasis workflows"
  - "Create internal policy memo on proxy usage and trust boundaries"
  - "Draft client-facing reference architecture"
  - "Define n8n-to-LangGraph escalation criteria"

Source Notes

Primary / official sources used in this artifact

Graph View