> For the complete documentation index, see [llms.txt](https://docs.helvia.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.helvia.ai/resources/release-notes/helvia.ai-release-2026.10.21.md).

# Helvia.ai Release 2026.10.21

21 October 2026

## 1. Connect MCP Servers and A2A Agents as Tools

You can now connect your agents to external MCP servers and A2A agents. An MCP server gives your agent ready-made actions, such as looking up an order or creating a ticket in your own systems. An A2A agent is another AI agent that yours can hand a question to and get an answer back. Both live under a new **Tools** category in **Workspace > Integrations**, and you decide which agents and which LLM v2 nodes can use each one.

**Why it matters:** Your agents can do more than answer from their own knowledge: they can act in your systems and draw on specialist agents. Managing both in one place, at the agent and node level, keeps you in control of what every agent can reach.

**Example use case:** A retail team registers its order-management MCP server and a partner's shipping agent. They make both available to the customer service agent, but not to the marketing agent. In that agent's workflow, they use them only in the LLM v2 node that answers "Where is my order?" questions.

**How it works:** Go to **Workspace > Integrations** and add an **MCP Server** (by its URL) or an **A2A Agent** (by its Agent Card URL). Then, in Designer, open the agent's **Tools** page to choose which MCP servers and A2A agents it can use, and check their connection status. Finally, in the agent's workflow, open an LLM v2 node and use its **Tools** tab to pick the ones that node can call. Each tool call appears in **Observatory > Interaction Logs**.

<img src="https://604830754-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FBM1xs3i59ajeTgi4uVfN%2Fuploads%2FHNQZAgHo4dXll5alzbLB%2Funknown.png?alt=media&amp;token=20ba7068-1007-46b3-98f7-677fdae35f22" alt="" width="563">

## 2. Use Your Own OpenAI-Compatible Models for Semantic Search and Session Analysis

The Semantic Search and Session Analysis plugins now work with OpenAI-compatible LLM integrations. You can run knowledge retrieval and conversation analysis on self-hosted or third-party models.

**Why it matters:** OpenAI-compatible integrations already worked with the LLM node. Now the same endpoint can index and search your Knowledge Base and analyze your chat sessions. This keeps your data inside your own infrastructure, which matters for organizations with strict data residency rules.

**How it works:** In your agent's **Plugins**, choose **OpenAI Compatible** as the provider for Semantic Search or Session Analysis, then select your integration and model.

## 3. Integration Secrets Are Now Write-Only

API keys, client secrets, tokens, and passwords saved in your integrations are no longer shown after you save them. This applies to every integration type and every Workspace member.

**Why it matters:** Your provider credentials stay protected from anyone with access to the console.

**How it works:** When you edit an integration, secret fields are always empty. Leave a field blank to keep the current value, or type a new value to replace it. If you change where an integration connects, such as its URL, endpoint, or tenant ID, you need to enter the secret again. OpenAI-compatible integrations now have an **Authentication** setting where you choose **None** or **API key**. If you need a key you saved earlier, retrieve it from your provider.

<img src="https://604830754-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FBM1xs3i59ajeTgi4uVfN%2Fuploads%2F3zdZKchOH1asoUO9R7HU%2Funknown.png?alt=media&amp;token=38ac76f1-8714-4221-9226-be3cfd037208" alt="" width="375">

## 4. Cleaner Interaction Logs by Default

Interaction Logs now hide variable changes by default, so LLM calls, HTTP requests, searches, tags, and redirects are easier to find. You can show variable changes again with one click.

**Why it matters:** Variable changes make up most rows in a typical log. Hiding them by default puts the events you usually look for at the top, so you spend less time scrolling when you debug a conversation. End-of-turn variable values are still available in the **Variables** tab.

**How it works:** This applies to **Observatory > Interaction Logs** and to the **Interaction Log** tab of a chat session. To see variable changes, open the **Type** filter and tick **Variable**. **Reset** in the filter returns to the new default.

## 5. Extract Parameters Fill In as You Write the Prompt

In the LLM v2 node, the **Extract Parameters** list now fills in as soon as you finish writing your prompt. You see the parameters right away, without closing and reopening the node.

**Why it matters:** When your prompt asks the model to return JSON, the node reads the keys from your example and creates matching parameters for you. Seeing them immediately lets you check and adjust them while you are still working on the prompt.

**Example use case:** A designer writes a prompt that returns sentiment, urgency, and a summary as JSON. When they click outside the prompt editor, three parameters appear with matching variable names, ready to use later in the workflow.

**How it works:** Set **Response Format** to **JSON Object** and include a JSON example in your prompt. The list fills in when you click outside the prompt editor or switch the format to JSON Object. Parameters you define yourself are never changed or overwritten, because the list is only filled when it is empty.

<img src="https://604830754-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FBM1xs3i59ajeTgi4uVfN%2Fuploads%2FNnEb0VtUZydoSbmISBn8%2Funknown.png?alt=media&amp;token=4f3b8c6b-0868-4620-ac74-856a1803f22e" alt="" height="134" width="466">
