Tess AI
AIRun agents and generate completions and embeddings on Tess AI by Pareto, the Brazilian multi-model platform. Agents work across model providers behind one connection.
What This Integration Enables
Tess AI, built by Pareto, is a Brazilian multi-model platform where the unit of work is an agent, not a prompt. Somebody in the business configures the agent, picks its model, writes its input questions and links its reference files, and then it exists as a thing other people can run. That is a genuinely different starting point from an API you send messages to, and it is the case FlowRunner was built for. We do not want to replace the agent your team already tuned. Orchestration as a Service means deciding when that agent runs, what context it gets, what happens to its output, and who has to look at it before anything irreversible happens next.
The connector gives an agent the full loop. List Agents and Get Agent discover what exists and, crucially, return the agent's configurable input questions, so a flow can learn which fields a given agent expects instead of hardcoding them. Execute Agent runs it, either waiting for the output inline or queueing the run and letting go. When you queue it, the On Agent Execution trigger fires on completion and FlowRunner registers and removes the underlying Tess webhook for you, so the asynchronous shape costs nothing to set up. Files can be uploaded and processed for retrieval, attached per run, or linked to an agent permanently as standing context. There is also an OpenAI-compatible completion endpoint served through an agent, which means existing OpenAI-shaped payloads work unchanged. Model slugs are free text, because the valid set depends on the target agent's configured model list and Tess publishes no live models endpoint, so a flow should resolve the agent before it assumes a model.
Without FlowRunner
With FlowRunner
Use Case Scenarios
Long-running agent work that does not block anything
A client brief arrives. The agent calls Get Agent to read the input questions the target Tess agent expects, maps the brief onto those fields, uploads the supporting research with Upload File and processing enabled, and calls Execute Agent with Wait For Completion turned off. The flow ends there. Minutes or longer later, On Agent Execution fires, a second flow picks up the finished output with the execution record already attached, and the deliverable is filed and posted to Slack. No step sat open waiting, and nobody wrote a polling loop.
One brief, several agents, one comparison
An account team keeps separate Tess agents for different content types. The flow runs List Agents filtered by type, dispatches the same brief to two or three of them in parallel with Execute Agent, and collects the responses with Get Agent Response as each one lands. The outputs are laid side by side in a Notion page with the agent name, the model slug used and the run time on each, and the account lead picks. The comparison is the product. Running three agents is only useful if somebody sees all three answers in one place.
Client reference material managed as standing context
When a new client is onboarded in HubSpot, the flow uploads their brand guide, tone document and product sheet with Upload File, confirms indexing finished with Get File, and links them to that client's agent so every future run reads them without anyone attaching anything. List Agent Files makes the standing context visible, which is the part teams usually cannot answer about their own agents. When the client updates their brand guide, the flow uploads the new version and proposes the swap rather than performing it.
Human-in-Loop Highlight
Delete Agent File looks like housekeeping and behaves like a config change to a shared system. Unlinking a file from an agent does not delete the file, but it silently changes what that agent knows on every future run, for every person and every flow that uses it, with no error and no notification. Somebody asks a question next week, gets a worse answer, and there is nothing in the output to explain why. Delete File is the harder version of the same thing: it removes the file and the indexed content derived from it permanently, and any agent that referenced it simply stops being able to read it. So FlowRunner treats a change to standing context as a change to a shared asset. The agent assembles the case and stops: "Removing the 2025 brand guide from the Acme content agent will change the output of every run using it. Two other flows call this agent. Replace with the 2026 version, remove without replacement, or cancel?" The account lead answers. This is human-in-the-loop placed where the damage is quiet rather than loud, which is usually where it belongs. The same shape covers Delete Webhook, where the failure mode is an external system that stops receiving callbacks and nobody noticing for a week.
Agent Capabilities
16 actionsAgents
4- List Agents Lists the agents available in the workspace, with free-text search across titles and descriptions and filtering by agent type. Results are paginated at 15 per page by default. Used to resolve an agent ID before executing it or managing its files.
- Get Agent Retrieves a single agent by ID, including its type, slug and the full list of configurable input questions it accepts. Used to learn which extra fields the agent expects, so a flow maps a brief onto them instead of guessing.
- Execute Agent Runs an agent with a user message and returns its execution record. Leave Wait For Completion enabled to receive the finished output in the same call, or disable it to queue the run and collect the result later. Attach previously uploaded files and memory collections to give the agent extra context for this run only.
- Get Agent Response Retrieves one agent execution by its response ID, including the current status and generated output. Used to collect the result of a run that was started without waiting.
Chat
1- Create Chat Completion Generates a chat completion through an agent using Tess AI's OpenAI-compatible endpoint, so existing OpenAI-shaped payloads work unchanged. The model slug must be enabled for the target agent. Streaming is not exposed, so one complete response is returned.
Files
7- List Files Returns a paginated list of the files stored in the workspace. Used to pick file IDs to attach to a run or to link to an agent permanently.
- Upload File Downloads a document from a URL and uploads it to the workspace, returning the file ID. Enable processing to have Tess index the file for retrieval immediately, or upload it raw and process it later.
- Get File Retrieves one stored file with its name and current processing status. Used to confirm indexing finished before referencing the file in a run.
- Process File Starts indexing a previously uploaded file so its contents become available to agents for retrieval. Used when the file was uploaded without immediate processing, or to reprocess after a replacement.
- Delete File Permanently removes a file from the workspace along with any indexed content derived from it. Agents that referenced the file lose access to it, which is why flows propose this rather than run it.
- List Agent Files Lists the files permanently linked to a specific agent, which it consults on every run without them being passed per execution. Used to answer what an agent actually knows.
- Delete Agent File Unlinks a file from an agent so it stops being used as standing context. The file remains in the workspace and can still be attached to individual runs, but every future run of that agent changes silently.
Webhooks
4- List Webhooks Returns a paginated list of every webhook registered in the workspace, across all agents. Used to audit which external endpoints currently receive execution callbacks.
- List Agent Webhooks Lists the webhooks registered for one agent, showing the callback URL, HTTP method and whether each subscription is active.
- Create Agent Webhook Registers an external callback URL that Tess AI calls when the specified agent finishes an execution. Used to notify a system outside FlowRunner. For FlowRunner flows, the On Agent Execution trigger manages its own subscription instead.
- Delete Webhook Permanently removes a webhook subscription so Tess AI stops sending execution callbacks to its URL. The agent and its past runs are unaffected, but anything downstream of that callback goes quiet.
Triggers
1 triggersEvent Triggers
1- On Agent Execution Fires whenever the selected Tess AI agent finishes an execution. FlowRunner registers the webhook on the agent automatically and removes it when the trigger is deleted, so no manual webhook setup is needed. Pair this with Execute Agent using Wait For Completion turned off to handle long-running agent work without holding a step open.
Frequently Asked Questions
What can FlowRunner do with Tess AI?
FlowRunner agents can run List Agents, Get Agent, and Execute Agent in Tess AI, plus 13 more actions.
Does connecting Tess AI to FlowRunner require OAuth?
No. Tess AI connects to FlowRunner with an API key, no OAuth flow required.
Can Tess AI trigger a FlowRunner workflow automatically?
Yes. Tess AI supports 1 trigger that can start a FlowRunner workflow automatically.
Start building with Tess AI
$100 in credits. No card required. Connect in minutes.