Got a question?
Before posting in the channel, please:
Read the full product documentation for this area.
Check the questions and answers listed below — your question may already be answered.
Search the channel history for similar threads.
If it's still unanswered, post it in the Researcher Teams channel — include what you tried and what you expected to happen.
Frequently asked questions
Getting started
Q: How do I get access to Evo Researcher? A: An org admin enables Evo Researcher via the Evo Platform Administrator tool. Once enabled, users are assigned either a User or Admin role. Researcher then appears in the Evo navigation menu automatically. No individual LLM key setup is required for internal Access users — the key is configured at the org level.
Q: What is the difference between a Chat and a Project? A: Chats are lightweight, single-user conversations for quick, one-off questions. They do not support folders, collaborators, or mid-session configuration changes. Projects are full workspaces — they support file uploads, connectors, skills, collaborators, and multiple conversations all sharing the same context. You can convert a Chat into a Project at any point using the three-dot menu next to the chat title in the left-hand navigation and selecting "Convert to project".
Q: Can I start a question in Navigator and continue it in Researcher? A: Yes. From Quick Asks in Navigator, there is an option to open the conversation in Researcher.
Q: Does Researcher work on mobile? A: The current UI is optimised for desktop. A more mobile-friendly experience is on the roadmap but is not available in the current release.
Q: Where should I go to build an AI agent or assistant? A: Use the Assistant Builder app (also called Agents Builder or CoWork Creator — these are consolidating into one product). Do not try to build agents within Researcher itself.
Connectors
Q: Can I use actions and connectors in a single chat? A: Yes, but only if they are configured as part of a template. From a single chat you can select a template and/or attach skills — the actions and connectors come in via the template. You cannot add connectors or actions directly to a standalone chat outside of a template.
Q: What connectors are available in Researcher? A: Researcher supports four types of connector: Evo product connectors (live data from other Access products), third-party connectors (e.g. OneDrive via OAuth), Agent Builder agents, and custom MCP servers. The specific Evo product connectors available depend on your region and which products your organisation has enabled. Check the Connectors tab in your project to see what is live for you.
Q: How does the AI use a connector? A: Adding a connector gives the AI a tool it can call when a message requires that data. The AI decides when to use it based on what you ask. The tool call is visible in the chat response so you can see what was queried. Adding more connectors than you need is not recommended — each one adds to the AI's decision space and can lead to less accurate responses.
Q: Is Researcher available as a node in Evo Workflow Automation? A: Yes. Evo Researcher (shown as "Evo Teamwork" in the EWA UI) is available as both a workflow node and as an MCP-based AI tool within the LLM step. This means EWA can read from and write to Researcher projects as part of automated workflows. To find your Project ID, ask Researcher directly in chat.
Q: Is a Databricks connector available for Researcher? A: Yes. Databricks is available as a third-party connector within Researcher.
Q: How do I invoke Researcher from an EWA workflow programmatically? A: POST to api/projects/{id}/chats/stream. You can find the correct chat/project ID by inspecting network traffic in your browser's developer tools while using Researcher.
Artifacts
Q: What types of artifacts can Researcher generate? A: Researcher can generate HTML pages, Markdown documents, PDFs, Excel spreadsheets, PowerPoint presentations, and charts. Word (.docx) is not supported.
Q: Do generated artifacts include Access branding? A: Not currently. PDF, Excel, and PowerPoint are simple artifacts — they do not carry Access branding or advanced formatting. Branded and more complex artifact generation is on the enhancements list. If you need a branded output now, generate the content as HTML and use your browser's print function (print > save as PDF) to download it as a PDF. This preserves the visual formatting from the HTML render.
Q: How do I save a branded HTML artifact as a PDF while preserving the branding and styling? A: Open the HTML artifact in a new browser tab, right-click anywhere on the page, select Print, then Save as PDF. Exporting directly from Researcher loses the branding.
Q: How do I share an artifact with someone? A: Artifacts are behind authentication — a direct URL will not work for someone who is not logged in. The best options are: download the artifact and share the file directly, or add the person as a view-only collaborator on the project so they can access it in Researcher.
Q: Can I add someone to a project just to view artifacts? A: Yes. You can add collaborators at read-only access, which lets them see the project's chats and artifacts without being able to edit or add to it.
Q: The AI responded in chat instead of creating an artifact. How do I get it to produce one? A: Ask explicitly: "Save that as an artifact I can preview and download." The AI will then call the artifact creation tool and it will appear in the Preview and Artifacts tabs.
Q: Can I paste an artifact into Outlook or Word? A: Use the Copy action rather than selecting the preview text manually. Copy preserves formatting. Selecting text from the preview can pull in raw Markdown syntax which appears as clutter in Outlook.
Q: Does Researcher support artifact versioning? A: Not currently. Artifact versioning is on the product roadmap.
Skills and templates
Q: What is the difference between a Skill, a Template, and Instructions? A: Instructions are per-project guidance you write directly into a project's Instructions tab. A Skill is a reusable instruction block published to the org library — any user can attach it to any project. A Template is a full project blueprint that carries instructions, uploads, connectors, and actions. The typical progression is: write it in Instructions for one project, publish it as a Skill once you are using it repeatedly, and fold it into a Template if the workflow also requires the same uploads and connectors every time.
Q: Who can create Skills and Templates? A: Org admins can create and publish Skills and Templates for the whole organisation. Users can create personal Skills for their own use but cannot create templates.
Q: Skills are being loaded for every query, even ones unrelated to the skill topic, which fills the context window. Is this expected? A: This is a known limitation. All active skills are currently loaded into context on every query. A feature to load skills on demand (only when relevant) has been requested.
Access and roles
Q: What can an Admin do that a User cannot? A: Admins can create and manage org-level templates and skills, manage users, and configure org-wide settings. Users can create and manage their own projects, upload files, chat with the AI, and collaborate on shared projects.
Q: I am an org admin. Can I hide a system template my organisation does not need? A: Yes. Org admins can hide system templates (cross-organisation templates set by the product team) without deleting them.
Tokens
Q: How do I use less tokens? A: Try these token-saving best practices in Evo Researcher:
Prompts
Be specific and direct; vague or verbose prompts add extra rounds and tokens.
One task per message, bundle what you can ask upfront rather than sending follow-ups
Don't repeat context the AI already has, if a file was uploaded earlier, don't paste its content again
Instructions (project system prompt)
Keep them concise and focused on how Researcher should behave, not what to build; every word is resent on every request in that project
Avoid embedded examples or data, link to a file instead
Connectors
Only add connectors the project needs, every connected tool's schema is sent on every request whether used or not
Uploads
Upload once per chat, don't re-upload the same file
Prefer smaller, focused files; save lengthy PPT or Word docs as PDF before uploading, same accuracy, far fewer tokens
For large volumes of files (for example 600+ images), use a connector such as SharePoint rather than uploading directly
For documents needed across many chats, put them in project instructions or a data space rather than re-uploading
Artifacts
Request Markdown first, HTML is more expensive in tokens; only generate HTML where it's actually needed
Ask for updates to the existing artifact rather than a new one, work from it in a new chat instead of refining it endlessly in one long conversation
Troubleshooting
Q: I cannot see a connector I expected to find. Why? A: Connector availability is partly controlled by feature flags and product enablement per region. If a connector you expected to see is missing, check with your org admin or post in the Evo Assistant Builder Teams channel.
Q: Why did my chat/artifact disappear when I lost connection? A: If you lose connection mid-session in a Project, the current chat state/artifact may not be preserved. This is a known limitation.
Q: Researcher is showing a blank white screen. What should I do? A: This usually follows a deployment. Wait a moment and refresh the page. If the issue persists, report it in the Researcher Teams channel.
Q: I have an issue with a connector or an agent. Who do I contact? A: Use the Evo Assistant Builder Teams channel.
Q: Uploads are not working in a single chat. A: The file being uploaded must be named in the prompt, e.g. image.png, CustomerFeedback.xlsx.
Q: A chat is showing a "Memory Limit Reached" error. A: This error can appear on new or empty chats and is not always caused by a long conversation. The most likely cause is system prompt bloat. When a project or chat has multiple connectors attached, each connector's tool definitions and skill instructions are loaded into the context before the conversation begins. On projects with several connectors or skills, this can push the context close to or over the limit before a single message is sent.
Workarounds while a fix is in progress:
Remove connectors from the project that are not needed for the current task
Start a new chat within the same project
If the issue persists across all chats in a project, remove skills one at a time to identify which is contributing most
Q: LiteLLM is returning errors about model names — older keys use "claude-sonnet-4-5" but newer keys use "anthropic/claude-sonnet-4-5". How do I fix this? A: Remove the "anthropic/" prefix from the model name. Newer LiteLLM key configurations include the provider prefix, which conflicts with older calling conventions. Use just "claude-sonnet-4-5" (or whichever model version applies).
Q: I'm migrating from Claude desktop and I had an MCP server connected. I can't see where to add it in Researcher. A: Go to your project, open the Connectors tab, click + Add connector, and select Custom. Paste your MCP server URL and add any auth headers required. Researcher will probe the server on save and surface its tools to the AI.
Q: I'm migrating from Claude desktop and I used desktop extensions to connect local tools with a single click. I can't find that option in Researcher. A: Researcher doesn't support local desktop extensions. For cloud-based tools (OneDrive etc.), use a third-party connector under + Add connector. For locally-hosted tools, you'll need to expose them as an MCP server, then connect via the Custom connector type.
Q: I'm migrating from Claude desktop and I used Cowork to run autonomous multi-step tasks. I can't find anything equivalent in Researcher. A: There are two paths depending on your use case. For on-demand agentic tasks, add an Agent Builder agent as a connector in your project. For scheduled or triggered automation, the equivalent is Evo Workflow Automation (EWA), which can call Researcher as a step in a workflow.
Q: I'm migrating from Claude desktop and Claude could read files directly from my local file system. I can't do that in Researcher. A: Researcher doesn't have direct file system access. Upload files to your project via the Uploads tab (30MB max per file).
Q: Where do I report bugs or give feedback? A: Use the Researcher Teams channel.