Langdock
AIRun workspace agents and generate completions and embeddings on Langdock, the GDPR-compliant enterprise AI platform. Agents work under a data residency posture legal has already signed off on.
What This Integration Enables
Langdock is what a company buys when the question is not "can our people use AI" but "can we show who did." The vendor positions it as a GDPR-compliant enterprise AI platform, and the connector reflects that: alongside the agent and completion surface there is a workspace user management API and a paginated audit log API, each gated behind its own API key scope. That last detail matters more than it sounds. A key with AGENT_API cannot read audit logs, and a key with AUDIT_LOG_API cannot invite anybody. FlowRunner builds on that separation instead of flattening it, which is the practical form of a Code of Conduct for agents: the credential itself limits what an agent is able to attempt.
On the working side, an agent can hold a conversation with a persistent workspace Agent that a person configured in Langdock, or define a temporary one inline for a single request when the task does not deserve a permanent object. It can create Agents programmatically, attach files for a single conversation or upload them into a Knowledge base for good, and run semantic search across every Knowledge base the key can reach, with results re-ranked and reduced to the strongest chunk per document. Completions and embeddings name their region, EU or US, on the call. Two operational limits are worth designing around: Langdock cuts off a non-streaming agent run after 100 seconds, so long tool-using work belongs in an asynchronous shape, and Knowledge base uploads process asynchronously, so a flow polls List Knowledge Base Files until syncStatus reads SYNCED before it searches.
Without FlowRunner
With FlowRunner
Use Case Scenarios
Contract questions answered from the Knowledge base, with the source
Legal operations keeps executed contracts in a Langdock Knowledge base. When a question arrives in Microsoft Teams, the agent runs Search Knowledge Bases, which returns the highest-scoring chunk per document after an LLM re-rank, then answers with Run Temporary Agent grounded in those chunks. Because the search step names its sources, the answer arrives with the contract it came from rather than as an assertion. New contracts are added with Upload File To Knowledge Base, and the flow waits on syncStatus before it tells anyone the document is searchable.
An internal Agent invoked as a step, not a destination
Someone in the business has already built and tuned an Agent in Langdock, with its instructions, its model and its linked Knowledge bases. FlowRunner treats that Agent as a callable step. A ticket arrives in Zendesk, the flow uploads the customer's attachment with Upload Agent Attachment, calls Chat With Agent with a JSON-schema-constrained structured output so the reply comes back as fields rather than prose, and writes those fields onto the ticket. The person who owns the Agent keeps owning it. FlowRunner decides when it runs, what context it receives and who checks the result, which is the whole point of sitting above agent builders rather than competing with them.
Audit evidence collected on a schedule
A compliance flow runs weekly. It walks List Audit Logs with a date window, pages through the cursor at 50 entries a time, and files the result to Google Sheets or a warehouse table so the evidence exists before anyone asks for it. Entries showing an actor who no longer appears in the HR roster, or activity outside expected hours, are separated into a review list and sent to the compliance owner. The routine collection is automated. The judgment about what an anomalous entry means is not.
Human-in-Loop Highlight
Invite Workspace Users is where this connector stops being a writing tool and starts being an administrative one, and it is the action FlowRunner puts a person in front of. The call takes a list of email addresses and an optional role, it silently skips existing members, and it auto-approves pending join requests, which makes it wonderfully safe to re-run and genuinely dangerous to run unattended. An agent syncing a joiners report should absolutely create Member invitations on its own. It should never decide that someone becomes an Admin, because an Admin in Langdock can reach the workspace's Knowledge bases and the account's user management, and revoking the role afterwards does not undo what was read in the meantime. So the flow splits the batch. Members go out. Everything requesting Editor or Admin stops and reaches the workspace administrator with the requester, their manager, the stated justification and their recent audit log activity in one message: "Three elevated grants requested this week. One is a returning contractor whose previous access was revoked in March. Approve individually?" That pause is FlowRunner's digital andon cord applied to access rather than to output, and it is the reason human-in-the-loop belongs in an AI platform integration at all.
Agent Capabilities
13 actionsAgents
6- Chat With Agent Sends a message to an existing Langdock Agent and returns its completed reply. The Agent must be shared with the API key. Supports attaching previously uploaded files by ID and requesting JSON-schema-constrained structured output, so the reply arrives as fields. Streaming is disabled, and long tool-using runs can exceed Langdock's 100 second server limit.
- Run Temporary Agent Runs a one-off agent defined inline for a single request without saving it to the workspace. Supply instructions, a model and a temperature, and optionally ground the reply in existing Knowledge bases. Used when the task is specific enough that a permanent Agent would be clutter.
- Get Agent Retrieves the full configuration of an existing Agent, including instructions, model, creativity, enabled capabilities, linked Knowledge bases and conversation starters. Requires the AGENT_API scope. Used to inspect what an Agent will actually do before invoking it in production.
- Create Agent Creates a persistent Agent in the workspace and shares it with the calling API key so it can be used immediately. Configures the system prompt, model, creativity, web search and image generation capabilities, and attaches Knowledge bases or uploaded files.
- Upload Agent Attachment Uploads a file fetched from a URL and returns a UUID that can be passed to Chat With Agent, Run Temporary Agent or Create Agent. Used when a document should be read for one conversation rather than permanently indexed.
- List Agent Models Lists every model available to the Agent API in the workspace, with each model ID, the region it is served from and whether it supports extended thinking. Used to resolve a valid model before creating or running an agent.
Completions and Embeddings
2- Create Chat Completion Generates a chat completion with an OpenAI model hosted inside Langdock's EU or US region, keeping the request within the chosen data boundary. Accepts an optional system prompt and returns the standard OpenAI chat completion payload. Streaming is disabled, so one complete response comes back.
- Create Embedding Converts one or more text strings into embedding vectors using an OpenAI embedding model hosted in the EU or US region. Each input stays under 8192 tokens. Used for semantic search, clustering and retrieval flows that need to name their region.
Knowledge Bases
3- Search Knowledge Bases Runs a semantic vector search across every Knowledge base shared with the API key and returns the most relevant chunks ranked by similarity. Results are re-ranked by an LLM and reduced to the highest scoring chunk per document, which makes this the retrieval step of a grounded answer flow. Requires the KNOWLEDGE_FOLDER_API scope.
- List Knowledge Base Files Lists every file in a Knowledge base with its MIME type, page count, auto-generated summary and processing state. Poll the syncStatus field to confirm a freshly uploaded document has finished embedding before searching it. Requires the KNOWLEDGE_FOLDER_API scope.
- Upload File To Knowledge Base Downloads a document from a URL and uploads it into a Knowledge base so agents and the search action can reach it. Supports PDF, Word, plain text, Markdown, HTML and PowerPoint up to 256MB, and rejects executables. Processing is asynchronous.
Administration
2- Invite Workspace Users Sends workspace invitations to a list of email addresses and optionally assigns each a role. Existing members are skipped and pending join requests are auto-approved, so the call is safe to re-run. Addresses that fail deeper validation come back in invalidEmails without failing the whole call. Requires an admin-created key with the USER_MANAGEMENT_API scope, and elevated roles belong behind an approval.
- List Audit Logs Retrieves workspace audit log entries recording who performed which action, when, and from where. Supports cursor-based pagination up to 50 entries per page and filtering by date window, entity type or actor. Requires the AUDIT_LOG_API scope for the matching workspace.
Frequently Asked Questions
What can FlowRunner do with Langdock?
FlowRunner agents can run Chat With Agent, Run Temporary Agent, and Get Agent in Langdock, plus 10 more actions.
Does connecting Langdock to FlowRunner require OAuth?
No. Langdock connects to FlowRunner with an API key, no OAuth flow required.
Can Langdock trigger a FlowRunner workflow automatically?
Langdock doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger.
Start building with Langdock
$100 in credits. No card required. Connect in minutes.