Eden AI
AIReach hundreds of AI models from dozens of providers through one Eden AI key. Agents run OCR, moderation, translation, and entity extraction through the Universal AI endpoint, plus chat completions, embeddings, and image generation through an OpenAI-compatible surface with automatic fallbacks.
What This Integration Enables
FlowRunner is a [BYOK](/concepts/byok/) platform, and Eden AI is that idea taken to its logical end: one key you own, reaching hundreds of models across dozens of providers. Instead of wiring a separate integration for every AI capability, agents get OCR, text moderation, translation, entity extraction, speech synthesis, and transcription through one connector, plus chat completions, embeddings, and image generation through an OpenAI-compatible surface that falls back to a backup model when the primary fails. Run Universal AI is the escape hatch that reaches every feature Eden AI aggregates, so a new capability on their side does not wait for a new connector on ours. There is a discipline question buried in this convenience: every one of these calls spends real money against your Eden AI account, and the async operations produce results that live in that account until deleted. FlowRunner treats both as governed resources. Agents process, classify, and extract autonomously; batch spend and destructive cleanup wait for a [human-in-the-loop](/concepts/human-in-the-loop/) decision.
Without FlowRunner
With FlowRunner
Use Case Scenarios
The document intake line
A scanned invoice lands in a watched [Google Drive](/integrations/google-drive-api) folder. The agent stages it with Upload File, runs Extract Text From Document for OCR, and passes the raw text to Ask Eden AI with a prompt that returns vendor, date, and totals as JSON. Clean extractions file into [Airtable](/integrations/airtable) with the source document linked. Totals that disagree with the purchase order do not post anywhere; they route to a person with the extraction and the scan side by side. The AP team reviews exceptions instead of keying documents.
Moderation that holds the line without becoming the judge
Every new post in a community feed runs through Moderate Text. Content scoring above the NSFW threshold is held, and the moderation team gets the post with its category scores in [Slack](/integrations/slack). Everything that clears the bar gets classified with Extract Topics and tagged before publishing. The agent never publishes what the model flagged and never deletes it either. A person decides what happens to borderline content, which is the only defensible place for that decision to live.
Calls become searchable knowledge
A recorded call submits through Transcribe Audio, which returns a job ID because transcription runs asynchronously. The agent polls Get Async Job until the diarized transcript is ready, summarizes it with Ask Eden AI, and writes the summary to [Notion](/integrations/notion) with decisions and follow-ups pulled out. Once the transcript is confirmed persisted downstream, Delete Async Job cleans up the stored result. The recording archive turns into notes people actually search.
Human-in-Loop Highlight
Every Eden AI action is metered spend, and Run Universal AI is the action that can spend fastest, because it reaches every expert model behind one interface. So when an agent is pointed at a backlog, thousands of archived documents to OCR and extract, it does not start burning through the queue. It processes a small sample first, then posts the evidence to the operator in Slack: "Sample of 50 documents processed: 47 extracted cleanly, 3 need review. The full backlog is 3,900 documents at roughly the sampled per-document cost. Run the batch, adjust the model choice, or hold?" Only an approval starts the full run. The same discipline applies to Delete Async Job, which permanently removes a job's stored result: the agent verifies the output was consumed downstream before cleanup, and asks when it cannot confirm. Spending your provider budget is a plan a person approves, not a loop an agent discovers.
Agent Capabilities
20 actionsUniversal AI and Async Jobs
5- Run Universal AI Calls any synchronous Eden AI expert model through the single V3 endpoint. The escape hatch that reaches every aggregated feature without waiting for a dedicated action.
- Run Async AI Job Submits a long-running feature for background processing and returns a job ID immediately. Used for heavy media and document workloads.
- Get Async Job Retrieves an async job by ID, returning status while running and the result when done. The polling step in every async flow.
- List Async Jobs Lists submitted async jobs, newest first, with feature and status filters. Used to reconcile in-flight work.
- Delete Async Job Permanently removes a finished job and its stored result from your account. Cannot be recovered, so agents confirm the result was consumed before cleanup.
Text Intelligence
4- Extract Named Entities Identifies people, organizations, locations, dates, and quantities in text. Used to structure unstructured intake before routing.
- Moderate Text Scores text for unsafe content with an overall NSFW likelihood and per-category detail. Used to hold borderline content for human review.
- Extract Topics Extracts a text's recurring themes with importance scores. Used to classify and tag content beyond keyword matching.
- Translate Text Translates text using aggregated engines including DeepL, Google, Amazon, and Microsoft. Used for multilingual intake and output.
Documents and Files
2- Extract Text From Document Runs OCR over an image or PDF, returning machine-readable text with per-word bounding boxes where supported. The first step of document intake.
- Upload File Uploads a file to Eden AI's store and returns a file ID any Universal AI feature can consume. Used to feed private documents without public URLs.
Speech and Audio
2- Text To Speech Converts text into spoken audio through engines including ElevenLabs, OpenAI, Google, and Amazon. Used for voice output in notification flows.
- Transcribe Audio Submits audio or video for asynchronous transcription with speaker diarization. Paired with Get Async Job to collect the transcript.
LLM Surface
4- Chat Completion Sends a full conversation to any routed large language model through an OpenAI-compatible endpoint. Used for multi-turn reasoning steps with fallback models.
- Ask Eden AI Sends a single prompt and returns just the answer text. The simple path for extraction, summarization, and classification steps.
- Create Embeddings Turns texts into vectors for similarity search and clustering. Used to build retrieval steps inside flows.
- Generate Image Creates images from text using routed models such as OpenAI gpt-image, Google Imagen, Amazon Nova Canvas, and Stability. Used for content production flows.
Catalog Discovery
3- List Features Returns the live catalog of features and subfeatures, the map for building Universal AI model IDs. Used before wiring a new capability.
- Get Subfeature Info Describes one subfeature in full: execution mode, endpoint, and exact input fields. Used to validate a call before spending on it.
- List Models Lists the LLMs reachable through the OpenAI-compatible endpoints with context length and per-token pricing. Used to pick models on cost and capability.
Frequently Asked Questions
What can FlowRunner do with Eden AI?
FlowRunner agents can run Run Universal AI, Run Async AI Job, and Get Async Job in Eden AI, plus 17 more actions.
Does connecting Eden AI to FlowRunner require OAuth?
No. Eden AI connects to FlowRunner with an API key, no OAuth flow required.
Can Eden AI trigger a FlowRunner workflow automatically?
Eden AI doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger.
Start building with Eden AI
$100 in credits. No card required. Connect in minutes.