Langdock and WorkingAgents both operate in the enterprise AI platform space, but they solve different problems at different layers. Langdock is an AI adoption platform – it puts AI chat and agents in front of employees. WorkingAgents is an AI governance platform – it controls what those agents are allowed to do once they’re deployed.
Understanding where they overlap, where they compete, and where they complement each other reveals how the enterprise AI stack is stratifying.
Langdock: The AI Adoption Layer
Langdock is a Berlin-based platform founded in September 2023 (YC S23). It raised $3.5M total from General Catalyst and La Famiglia – and turned that into $15M ARR by December 2025, 100K+ monthly active users, and 4,000+ customers including Merck (33,000 MAU in a single account), The Economist, UNICEF, Personio, and GetYourGuide.
The growth numbers are exceptional. $1M to $15M ARR in 14 months on $3.5M in funding. That’s capital efficiency rarely seen outside of PLG SaaS.
What Langdock Does
Five products:
-
Chat – Multi-model AI chat interface for employees. Document upload, image generation, code execution, prompt library. The “give everyone AI access” product.
-
Agents – Custom AI chatbots with defined personas, instructions, and grounded knowledge. Deployable in Langdock, Slack, and Microsoft Teams. Subagent delegation launched February 2026.
-
Workflows – Multi-step automation combining AI, integrations, and custom logic. Visual builder with run history and debugging.
-
Search (Company Knowledge) – Unified search across SharePoint, Google Drive, Confluence, and connected platforms. Synthesized answers with source links.
-
API – Unified gateway to 40+ AI models with a 10% surcharge on provider token rates. Embedding API and Knowledge Folder API for RAG pipelines.
Langdock’s Strengths
- Model flexibility: 40+ models across Anthropic, OpenAI, Google, Meta, Mistral, and DeepSeek. EU-hosted and US-hosted options.
- European data sovereignty: ISO 27001, SOC 2 Type II, GDPR compliant. EU hosting on Microsoft Azure. Single-tenant, BYOC, and on-premise deployment options for enterprises above 1,000–5,000 seats.
- MCP support: Full MCP client support – agents can connect to any MCP server via Streamable HTTP and SSE. OAuth 2.0 authentication for MCP connections.
- Pricing clarity: EUR 25/user/month (Business) with included AI model credits. Enterprise pricing is custom.
- Integration breadth: Native connectors for Google Drive, Notion, Confluence, Slack, Salesforce, Jira, Asana, Microsoft Teams, SharePoint. Source-tool permissions mirrored 1:1 – if a user can’t access a file in Google Drive, the AI can’t either.
Langdock’s Gaps
- Coarse-grained access control: Permissions operate at the workspace and integration level, not at the individual tool or action level. There’s no public documentation of per-tool permission keys, per-agent identity, or fine-grained operation scoping.
- No detailed audit trail: No public documentation of structured per-action audit logging with cost attribution, guardrail evaluation records, or cross-system action chains.
- No behavioral guardrails: Actions require human confirmation, but there’s no documented pre/during/post execution guardrail system – no PII detection, no injection prevention, no content filtering at the platform level.
- European concentration: Customer base is heavily German/European. Limited US enterprise penetration documented.
- Assistance over execution: Workflows and agents are designed for employee augmentation. The platform doesn’t position itself as an autonomous agent orchestration layer.
WorkingAgents: The AI Governance Layer
WorkingAgents is an AI governance platform – the access control and orchestration layer between AI agents and the enterprise systems they interact with. Three gateways, one control plane:
What WorkingAgents Does
-
AI Gateway – Unified proxy to 250+ LLMs with smart routing by cost and latency. Automatic failover.
-
Agent Gateway – Control plane for multi-step agentic workflows. Retries, timeouts, fallbacks, and human-in-the-loop checkpoints. Prevents runaway loops – catches agents stuck in retry chains before they burn through API budgets overnight.
-
MCP Gateway – Centralized tool registry with per-user token management, permission boundaries, and guardrails on every tool call. Four-layer authentication: gateway, team, service, and custom.
WorkingAgents’ Strengths
-
Granular access control: Capability-based permissions at the individual tool and operation level. An agent can be granted
hubspot_read_contactswithouthubspot_send_email. Permission keys with TTL for temporary access. Per-agent identity – the system distinguishes between 10 different AI agents sharing infrastructure, not just API tokens. - Three-checkpoint guardrails: Pre-execution (input validation, injection prevention), real-time (behavioral enforcement, human approval), post-execution (PII detection across 20+ categories, credential masking, content filtering). Guardrails run on every tool call, not just at entry.
- Structured audit trails: Every action logged with user, agent, tool, arguments, permission check result, guardrail evaluations, latency, and cost. Cross-system audit chains – when an agent reads CRM, writes Slack, and updates a database, the full sequence is traced.
- Virtual MCP Servers: Configuration-driven permission boundaries per team, role, or use case. No code changes – pure configuration. Agents see only the tools they’re authorized to use.
- Self-hosted by design: Deploy in your VPC, data center, or air-gapped network. Zero data egress.
- OTP fault tolerance: Built on Elixir/OTP with supervision trees, per-user process isolation, and hot code reloading. Crash in one subsystem doesn’t affect others.
WorkingAgents’ Gaps
- No end-user AI chat product: WorkingAgents doesn’t provide a consumer-facing chat interface or employee-facing AI assistant. It governs agents – it doesn’t replace the need for an AI interaction layer.
- No visual workflow builder (yet): Workflow orchestration is API-driven, not visual. A low-code builder is on the roadmap.
- Earlier stage: Smaller team, pre-revenue compared to Langdock’s $15M ARR.
Head-to-Head Comparison
| Dimension | Langdock | WorkingAgents |
|---|---|---|
| Primary function | AI adoption – put AI in employees’ hands | AI governance – control what agents can do |
| User-facing? | Yes – chat, agents, search for employees | No – infrastructure for agents and admins |
| LLM support | 40+ models, 6 providers | 250+ models via AI Gateway |
| Access control | Workspace-level, integration-level | Per-tool, per-agent, per-operation |
| Permission granularity | Role/workspace scoping | Capability-based keys with TTL |
| Agent identity | Token-based | Individual per-agent |
| Guardrails | Human confirmation on actions | Pre/during/post execution on every call |
| PII detection | Not documented | 20+ categories, automatic redaction |
| Audit trails | Optional analytics | Structured per-action with cost attribution |
| MCP role | MCP client (consumes MCP servers) | MCP server (provides governed tools) |
| Deployment | SaaS, single-tenant, BYOC, on-prem | Self-hosted, VPC, air-gapped |
| Compliance certs | ISO 27001, SOC 2, GDPR | Designed for SOC 2, HIPAA, GDPR, FedRAMP |
| Pricing | EUR 25/user/month | Per-deployment (governance layer) |
| Target buyer | CHRO, COO, CTO (“AI for every employee”) | CISO, CTO, GRC (“govern every agent”) |
Where They Compete
Enterprise AI gateway
Both offer multi-model LLM access. Langdock’s API product routes to 40+ models with a 10% surcharge. WorkingAgents’ AI Gateway routes to 250+ models with cost/latency optimization. For enterprises that need a unified LLM proxy, this is direct overlap.
Agent building and deployment
Langdock’s Agents product lets teams build custom AI chatbots with knowledge grounding and deploy them in Slack and Teams. WorkingAgents’ Agent Gateway orchestrates agent workflows with governance. If an enterprise wants governed agent deployment, both platforms have a claim – Langdock for the agent itself, WorkingAgents for the guardrails around it.
Enterprise search / RAG
Langdock’s Company Knowledge product provides unified search across connected platforms. WorkingAgents’ MCP Gateway can expose knowledge bases as governed tools. Different approaches to the same enterprise need.
Where They Don’t Compete
Langdock owns the employee interface
WorkingAgents has no chat product, no agent builder UI, no employee-facing search. Langdock is the layer where humans interact with AI. WorkingAgents doesn’t try to be this.
WorkingAgents owns the governance layer
Langdock’s access control is workspace-level. It mirrors source-tool permissions but doesn’t add its own granular permission model on top. WorkingAgents’ entire value proposition is the per-tool, per-agent, per-action governance layer – capability-based keys, TTL grants, three-checkpoint guardrails, structured audit trails.
For regulated industries (finance, healthcare, legal) where every agent action must be auditable and permissioned, WorkingAgents provides infrastructure that Langdock doesn’t.
MCP: Client vs. Server
This is the most architecturally significant distinction. Langdock is an MCP client – it consumes tools exposed by MCP servers. WorkingAgents is an MCP server – it exposes governed tools for MCP clients to consume.
This means they’re not just non-competing in this dimension – they’re complementary. A Langdock agent could connect to a WorkingAgents MCP Gateway to access enterprise systems with per-tool permissions, guardrails, and audit trails that Langdock doesn’t natively provide.
The Complementary Architecture
The most interesting finding from this analysis is not competition – it’s fit.
Langdock excels at: getting AI into the hands of every employee, supporting 40+ models, providing a polished chat and agent interface, European data sovereignty.
WorkingAgents excels at: governing what agents do once deployed, per-action permissions, three-checkpoint guardrails, structured audit trails, self-hosted deployment.
The gap in Langdock’s platform is precisely what WorkingAgents provides. Langdock agents that need to access enterprise systems – CRM, ERP, ticketing, databases – could route through WorkingAgents’ MCP Gateway to get governed access with audit trails. Langdock provides the intelligence layer. WorkingAgents provides the control layer.
A Merck deployment with 33,000 monthly active users running AI agents across sensitive pharmaceutical data would benefit from both: Langdock for the employee-facing AI experience, WorkingAgents for the governance layer ensuring those agents only access what they should, every action is logged, and every high-risk operation requires approval.
Strategic Implications
For Langdock: As they move upmarket into regulated enterprises (finance, healthcare, government), the governance gap becomes a sales blocker. Enterprises in these sectors will ask about per-action audit trails, granular permissions, and compliance certifications that Langdock doesn’t currently offer. They’ll either build this layer or partner with someone who has it.
For WorkingAgents: Langdock’s 4,000+ customers and 100K+ MAU represent a distribution channel. If Langdock agents can connect to WorkingAgents’ MCP Gateway, every Langdock enterprise deployment becomes a potential WorkingAgents customer – not by replacing Langdock, but by sitting underneath it.
For enterprises: The AI stack is stratifying into layers: model providers (Anthropic, OpenAI), adoption platforms (Langdock, Glean, Dust), governance platforms (WorkingAgents), and enterprise systems (Salesforce, ServiceNow, SAP). The winners won’t be the ones that try to own every layer – they’ll be the ones that own their layer definitively and integrate cleanly with the rest.
WorkingAgents is an AI governance platform specializing in agent access control, orchestration, and security for enterprises deploying AI at scale.