AI, deploy, and more
Fix failing workflows with self-healing
When a published workflow fails in production, Duaer Designer investigates, tests a fix on a copy, and sends it to Reviews. The workflow does not change until a reviewer applies the fix.
What self-healing does
When a published workflow fails in production, Duaer Designer reads the failed execution, works out the cause and prepares a fix. It tests the fix on a temporary copy of the workflow, then sends the result to Reviews in the sidebar. Your workflow does not change until a reviewer applies the fix.
- Starts self-healing: runs started by a trigger, a webhook or a chat, on the current published version.
- Does not: manual test runs, evaluations, retries you start, sub-workflows called by another workflow, and error workflows.
- The failed execution must be saved. If the workflow is set not to save failed executions, there is nothing to investigate.
- A workflow has at most one investigation or one open result at a time.
Self-healing is on for every Duaer workspace: project settings have a Self-healing section and the Duaer console has a Self-healing tab under Personal. Nothing is investigated until a project adds a configuration.
Set it up in a team project
- Open the project, go to Project settings, and under Self-healing select Add configuration.
- Under Scope, choose All, or choose Select manually and add workflows.
- Add reviewers: one at a time, or choose the project members entry (People from, then the project name) to make every project member a reviewer.
- Keep Enable self-healing on and select Create configuration.
Project admins, and instance owners and admins, can set up self-healing in a team project. A project can have one configuration for all workflows and any number of configurations with selected workflows; a workflow can be selected in only one of them. A configuration with selected workflows comes before the one for all workflows, even while it is paused. A workflow moved to another project leaves its old selection, and later failures follow the new project's configuration.
Set it up in your personal project
In the Duaer console, switch to Personal beside the title, go to the Self-healing tab and choose the workflows to watch. In a personal project you are the only reviewer. Only the owner of a personal project can set it up, and instance owners and admins do not see results from someone else's personal project. People who only use chat cannot set up self-healing.
Reviewers
- Reviewers must be members of the project. In team projects, instance owners and admins count as members.
- To act on a result, a reviewer needs edit access to its workflow, from a project role, an instance role or the workflow being shared with them.
- With the project members entry, reviewers follow current membership: people who join the project later become reviewers too.
- Each configuration needs at least one reviewer who can edit a workflow it covers. Without one, self-healing pauses.
Self-healing sends no email or message. Results appear only in the reviewers' Reviews, so reviewers need to check there.
Review a result
Open Reviews in the sidebar and select a result. It shows the investigation report, Changes (the original and the suggested changes side by side), Activity, the failed execution, and Usage (credits, duration and tokens).
- Fix ready: a tested change you can apply.
- Needs attention: Duaer Designer needs something from you, often a credential to set up. Read the report, then continue.
- Could not fix: no fix was found and the result closed on its own. The report says what to try next.
Apply and publish saves the change and publishes the workflow; you need permission to publish it. Under Continue in…, Continue with Duaer Designer and Continue with Editor save the suggested change to the workflow draft without publishing, so you can finish it. Discard and close closes the result and leaves the workflow as it is. If the workflow changed after the suggestion, the result shows Outdated: look at the current workflow before you change it.
What Duaer Designer can and cannot do
- It reads the failed published version, executions you may see, data tables and node details.
- It changes only a copy: tests run on a temporary, unpublished copy that is deleted afterwards. Your workflow, its history and its published version stay as they are.
- As in any Duaer Designer test, some nodes call outside services while the copy runs.
- It cannot save, publish or set up credentials, and it cannot use MCP servers or a browser. When Duaer Designer is blocked from web search or opening URLs, so is self-healing.
- It changes only nodes and connections. When a credential is missing, the result is Needs attention, with the steps to set it up.
Credits
Each investigation spends your organization's AI credits, like any Duaer Designer conversation, and the result's Usage shows how many. To spend less, select fewer workflows or turn off Enable self-healing for a while.
When it investigates again
- While a workflow has an open Fix ready or Needs attention result, it is not investigated again.
- After you discard or close that result, the next production failure can start a new investigation.
- After you apply a fix, the workflow is investigated again only once it is published again and fails again.
- After Could not fix, the same published version is not investigated for 24 hours. Publishing a different version ends that pause.
- Turning off Enable self-healing stops new investigations and cancels running ones; existing results stay in Reviews. Deleting a configuration deletes its investigation history and results.
Questions
Does self-healing change my workflow by itself?
No. Duaer Designer tests a fix only on a temporary copy and sends the result to Reviews. The workflow changes only when a reviewer selects Apply and publish, or continues to the draft.
Who is told when a fix is ready?
Nobody gets an email or message. The result appears in the reviewers' Reviews, and reviewers need to check there.
Does self-healing cost anything?
Each investigation spends your organization's AI credits, like any Duaer Designer conversation. The result's Usage shows how many.
In this section
Build your first AI Agent in Duaer
Connect a chat model to an AI Agent in Duaer, then add tools when you need them. This page is the smallest workflow that runs.
AITurn Agents on in Duaer
In Duaer, Settings > Agents turns Agents on for the instance and shows the sandbox shared with Duaer Designer.
AIUse Chat hub in Duaer
Chat hub gives people a chat entry without the canvas. From Duaer v3 it is off by default. You can turn it on temporarily with DUAER_ENABLED_MODULES.
AIRetrieve relevant context with RAG in Duaer
Write documents into a vector store, then retrieve pieces for the model when someone asks. Duaer RAG usually has a load branch and a retrieve branch.
AIUse $fromAI for tool parameters in Duaer
In Duaer: $fromAI marks a parameter the AI Agent fills when it calls a tool. Keys may only use letters, numbers, underscores, and hyphens.
AIUnderstand why to evaluate in Duaer
In Duaer, use evaluations against known test cases so AI workflows stay reliable across inputs.
AIRun light evaluations in Duaer
In Duaer, during development run a workflow against a few known cases and compare outputs side by side before you add formal metrics.
AIUse metrics to measure quality in Duaer
In Duaer: Run an AI workflow against a test set and record scores you calculated, so you can compare versions over time.
AIFix common evaluation issues in Duaer
In Duaer, when triggers, field maps, or chat output fight the evaluation nodes, check the wiring first.
AITurn on the Duaer MCP server
In Duaer, expose the instance as an MCP server so outside clients operate workflows with permission. Turn access on in the instance first.
AIDraft Duaer workflows with AI
In Duaer, describe your goal in plain language. The builder generates or edits digital-organization nodes. You still confirm on the canvas.
AIConnect an MCP client to Duaer
After MCP access is on in Duaer, put the connection URL and auth into the client config.
AIUse Duaer Designer on cloud
Duaer Designer on cloud is toggled in Settings and helps edit workflows.
AIManage MCP access on Duaer cloud
After you turn on MCP access on a Duaer cloud instance, outside clients can connect.
DeploySet up Duaer Designer on a self-hosted instance
Self-hosted Duaer configures Duaer Designer in its setup wizard in Settings. Sandbox and web search are optional. Prepare environment variables before startup.
DeployHost Duaer with Docker
From Duaer v3, Docker is the only supported deployment. If the instance check finds you are not in a container, it opens this page.
DeploySet up task runners as an external process in Duaer
Duaer deprecates the internal task runner. Run the runner separately, set it to external, and share DUAER_RUNNERS_AUTH_TOKEN with the launcher.
DeployConfigure binary storage with environment variables in Duaer
In Duaer, use environment variables for binary storage mode and path. The main process and workers must share one configuration.
DeployConfigure security environment variables in Duaer
In Duaer, security switches such as SSRF are configured with environment variables. From v3 the default blocked IP ranges are wider, and the check links here.
DeployConfigure digital-organization environment variables in Duaer
In Duaer, environment variables can set behavior such as the default caller policy. If the default is still any, the v3 check reports it and opens this page.
DeployManage security policies in Duaer
In Duaer, the security page in Settings points docs here for instance-level security policy entry points.
MoreCreate and edit credentials in Duaer
In Duaer, credentials are stored on the instance and protected by the encryption key. Nodes reference a credential instead of putting secrets in parameters.
MoreUse end-user credentials in Duaer
In Duaer, end-user credentials belong to the end user’s own connections. At runtime the triggerer’s account is used, not one shared secret.
MoreConfigure digital-organization settings in Duaer
In Duaer, settings include timezone, error handling, and caller policy. After v3 removes “Any workflow”, the check links to the who-can-call section.
MoreConfigure an embedded deployment in Duaer
When you embed Duaer in another product, configuration can overwrite credentials and related behavior.
MoreOpen the courses entry in Duaer
In Duaer, the Courses link in the product opens this page. It is an entry to practice notes, not a separate video school.
MoreUnderstand how items flow in Duaer
In Duaer, this chapter explains what one input looks like and how a node runs with many items.
MoreLooping docs moved in Duaer
In Duaer, legacy URL. Per-item execution and Execute Once are on the flow-logic looping page.
MoreUse the Code node instead in Duaer
In Duaer, the Function node is removed. Write JavaScript in the Code node. This URL remains so old links are not empty.
AIHow Duaer Designer edits a workflow
Duaer Designer changes only the nodes you named, and it replies in the interface language selected in the sidebar. Attachments you reject leave the conversation.
AICheck and rate what Duaer Designer delivers
Duaer Designer checks a built workflow against acceptance samples you confirm, and calls it verified only when every sample matches. After delivery, rate the reply with thumbs up or down and say what went wrong.
AISchedule a task for a Duaer digital employee
In Duaer, a digital employee task can run hourly, daily, weekly, monthly, or on a cron expression. A custom cron shows a readable sentence and the next run time.
AIInstall verified community nodes with Duaer MCP
In Duaer, a connected MCP client can install a verified community node. The client needs that permission, and it asks you before it installs.
AIConnect Databricks Embeddings in Duaer
In Duaer, Embeddings Databricks is a sub-node on an AI node. It turns text into vectors.
AISet MiniMax output length in Duaer
In Duaer, MiniMax chat-model output length is Maximum Number of Tokens under the node’s Options. The ceiling follows the model you select.
AI recipesConnect Claude to Duaer with MCP
Turn on the Duaer MCP server and add a custom connector in Claude, so Claude can find, run, and build your Duaer workflows.
AI recipesBuild a Duaer chatbot that answers from Google Drive docs
In Duaer, chunk the documents in a Google Drive folder into a vector store, then let an AI Agent retrieve the relevant pieces to answer each question.
AI recipesLet AI draft Gmail replies to new emails with Duaer
In Duaer, each new email goes to a model that writes a reply, saved as a Gmail draft in the same thread. You review and send; the AI never sends for you.
AI recipesUse DeepSeek models in a Duaer workflow
In Duaer, add your DeepSeek API key once, then plug deepseek-chat or deepseek-reasoner into an AI Agent or Basic LLM Chain.
AI recipesRun local models with Ollama in Duaer
A self-hosted Duaer can reach Ollama on the same machine or network and plug open models such as Llama and Qwen into AI Agents and RAG, keeping data on your servers.
AI recipesAuto-reply to WeCom app messages with AI in Duaer
In Duaer, when a member messages a WeCom self-built app, a model writes the reply and Duaer sends it back as an app message, optionally with memory.
AI recipesLet Claude read and write Google Sheets through a Duaer MCP server
In Duaer, expose Google Sheets read and append as MCP tools with Duaer Server Trigger, so Claude can look up and add rows on request.
AI recipesTranslate one Excel column in bulk with AI in Duaer
In Duaer, upload a spreadsheet, translate one column row by row with Basic LLM Chain, merge the translation back, and export.
MCPConnect AI tools to Duaer
Connect Claude, Claude Code, ChatGPT, Cursor, VS Code and Codex to the Duaer MCP server: the URL, how sign-in works, steps for each tool, scopes and common problems.