--- title: "TextCortex Integration" description: "Draft, rewrite, and summarize with the TextCortex API, an OpenAI-compatible text generation service. Agents produce copy inside the workflow that already holds the customer context." url: https://flowrunner.ai/integrations/textcortex-ai date_modified: 2026-08-07T17:20:23-07:00 --- # TextCortex [AI](https://flowrunner.ai/integrations/category/ai-llms) Draft, rewrite, and summarize with the TextCortex API, an OpenAI-compatible text generation service. Agents produce copy inside the workflow that already holds the customer context. [Verified](https://flowrunner.ai/integrations/verified "What does verified mean?") · 5 actions · API key · available [TextCortex website](https://textcortex.com/) · [Platform Documentation](https://docs.textcortex.com) · Capability data verified 2026-08-04 1. A quarterly review flags forty support macros as written before the product was renamed 2. Agent calls Get Credit Balance to confirm the account can cover the whole batch, so the run does not stop halfway 3. Agent pulls each macro's current text and the tickets it was most recently used on 4. Agent runs Create Response with the tone and terminology rules as Instructions and the existing macro as Input 5. Agent posts the old and new text side by side to the support leads channel 6. The support lead approves the rewrite before anything is written back over the original macro ## What This Integration Enables TextCortex is a writing API, and on the surface that puts it in a crowded room. The thing that actually separates it in a workflow is Get Model, which returns the deployment jurisdiction a model is served from: EU, EEA, US, UK, CH or GLOBAL. That is a field almost nobody else exposes, and it changes what an agent can be trusted to do. A flow can resolve a model, read where it runs, and decide whether this particular piece of text is allowed to go there, rather than relying on a decision someone made in a config file eighteen months ago. Combined with List Models reading live so newly released models appear without a connector change, it means model ids are never hardcoded and jurisdiction is never assumed. The generation surface is deliberately small and well split. Create Chat Completion takes a full message list, so it carries system instructions, multi-turn context and assistant priming, and it is the right choice when the conversation history matters. Create Response takes reusable guidance in Instructions and the material to work on in Input, which is the better shape for the far more common case of "apply these rules to this text" run across hundreds of items. Get Credit Balance is the pre-flight check that keeps a large batch from failing in the middle. Five [connectors](https://flowrunner.ai/concepts/connectors) worth of capability would be generous; five well-chosen actions that cover drafting, rewriting, discovery, residency verification and cost control is more useful than fifty that do not. ### Without FlowRunner **Macro library ages quietly**: Saved replies keep the old product name until a customer points it out **Rewrites done one at a time**: Whoever notices the problem fixes the one macro in front of them **Jurisdiction assumed**: Nobody checks where the model serving a rewrite is actually deployed ### With FlowRunner **The whole library reviewed in one pass**: Every macro is rewritten against the same tone and terminology rules **Originals survive the rewrite**: The new text is proposed next to the old, and the overwrite waits for a person **Jurisdiction verified per model**: Get Model reports the deployment region before any customer text is routed to it ## Use Case Scenarios ### A macro library rewritten in one pass, applied one approval at a time Support macros drift. Product names change, tone guidance is updated, and the saved replies stay as they were. The agent pulls every macro from [Zendesk](https://flowrunner.ai/integrations/zendesk) or [Help Scout](https://flowrunner.ai/integrations/help-scout), checks Get Credit Balance so the batch cannot die halfway through and leave the library half-modernised, and runs Create Response over each one with the tone rules as Instructions and the existing macro as Input. The results are posted as a diff. Nothing is written back until a support lead says so, because a macro library where half the entries were rewritten and half were not is worse than one that is uniformly out of date. ### A first draft reply that a human finishes A ticket arrives. The agent gathers the customer's history, the relevant knowledge base article and the account tier, and calls Create Chat Completion with that context as system instructions and the ticket thread as the message list. The draft is attached to the ticket as an internal note, never sent. The agent handles the assembly work that makes a good reply possible, which is where most of the time actually goes, and the person handles the sentence that reaches the customer. ### Routing text by where the model runs Before any flow sends customer text for processing, it calls Get Model on the configured model and reads the deployment jurisdiction. Text classified as containing personal data routes only to models reporting EU or EEA; general marketing copy can use anything. When a model is retired and List Models returns a replacement, the flow re-checks the jurisdiction rather than inheriting the old assumption. The check is two calls and it turns a policy statement into something the flow actually enforces. ## Human-in-Loop Highlight Generating a rewrite is cheap and reversible. Writing it over the original is neither. That is the whole gate on this connector, and it sits later in the flow than people expect. When an agent rewrites forty support macros, forty product descriptions or a library of email templates, the generation step is safe to run unattended: worst case, you discard the output and spend a few credits. The write-back is the irreversible half, because the source text in the helpdesk is usually the only copy, and once it is overwritten, the phrasing that a team spent two years refining against real customer confusion is gone. So the agent produces the whole batch, posts it as a before-and-after list, and asks the support lead: "Forty macros rewritten against the new terminology rules. Twelve changed only the product name. Six changed the meaning of the escalation path and are flagged. Apply all, apply the twelve trivial ones and hold the rest, or discard?" The lead answers, and only then does anything get overwritten. The agent does the labour; the person owns the destruction. That is [human-in-the-loop](https://flowrunner.ai/concepts/human-in-the-loop) at the point where it is actually load-bearing rather than at the point where it is easiest to add. Agent processes routinely Detects exception requiring judgment Clear match Continues automatically Ambiguous Routes to human via email Human decides Agent resumes with decision ## Agent Capabilities 5 actions ### Text Generation 2 - **Create Chat Completion** Generates a reply from a full conversation history using any TextCortex-hosted model. This is the OpenAI-compatible endpoint and accepts a message list of system, user and assistant turns, so it supports system instructions, multi-turn context and assistant priming. Streaming is disabled, so the complete answer returns in one response. - **Create Response** Generates content from a single instruction plus input text using the OpenAI-compatible Responses endpoint. Simpler than Create Chat Completion when there is no conversation to carry: reusable guidance goes in Instructions, the material to work on goes in Input. Server-side response storage is not supported, so the full output always returns inline. ### Models 2 - **List Models** Returns every TextCortex model currently reachable through the public API, with its id, owner and release timestamp. Read live, so newly released models appear without a connector change. Used to resolve a valid model id before generating, and to audit what an account can reach. - **Get Model** Retrieves metadata for a single model, including its owner, release timestamp and the deployment jurisdiction it is served from: EU, EEA, US, UK, CH or GLOBAL. Used to verify data residency before routing text to a model, which is the check most writing APIs give you no way to perform. ### Account 1 - **Get Credit Balance** Returns the remaining API credit balance and its currency for the configured key. Called before a large batch to fail early, or on a schedule to raise an alert while there is still time to act on it. ## Frequently Asked Questions ### What can FlowRunner do with TextCortex? FlowRunner agents can run Create Chat Completion, Create Response, and List Models in TextCortex, plus 2 more actions. ### Does connecting TextCortex to FlowRunner require OAuth? No. TextCortex connects to FlowRunner with an API key, no OAuth flow required. ### Can TextCortex trigger a FlowRunner workflow automatically? TextCortex doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger. **Work at TextCortex?** This integration exposes TextCortex to AI agents on every FlowRunner plan, including through MCP, at no cost to you. [See what FlowRunner offers integration partners](https://flowrunner.ai/integrations/partners), including how to keep this page current. --- Markdown version of https://flowrunner.ai/integrations/textcortex-ai. Site index: https://flowrunner.ai/llms.txt