Most Asked Questions
These are the questions users typically ask after reading the docs and starting to use Arkios in real workflows.
Getting Started and Setup
I read the docs. What should I do first in the product?
Start with one focused use case: create one agent, test it in chat, then iterate.
Use this sequence:
What is the fastest way to get a reliable first agent?
Keep scope narrow. Pick one job, add clear instructions, connect only relevant memory, and enable only required tools.
Should I start with chat first or memory first?
Start with chat for baseline behavior, then add memory to improve factual consistency.
Read: Memory System.
How do I know if I should use a custom agent or general AI chat?
Use general chat for quick, ad hoc prompts. Use custom agents when you need repeatable behavior, team consistency, memory grounding, and controlled tools.
Read: Agents Overview.
Agents
My agent gives broad answers. How do I make responses more specific?
Tighten the system prompt, define output format, and limit the scope of allowed sources/tools.
Read: Create Agents.
How many tools should I attach to a new agent?
Start with the minimum set. Too many tools increase noise and unpredictable behavior.
Read: Tools & Connectors.
A tool is not showing up in my agent. What usually causes this?
Most commonly:
- The tool is not enabled in Tool Hub
- Auth scope is misconfigured
- Access settings prevent the current user from using it
Read: Tool Hub.
How do I keep agents from answering outside policy?
Add explicit boundaries in the system prompt and require the agent to state uncertainty when sources are missing.
What should I test before sharing an agent with my team?
Test real prompts from your team, edge cases, and failure conditions. Verify tool behavior, source grounding, and output format consistency.
Where do I configure an agent to use company knowledge?
Add memory collections in knowledge settings for that agent.
Read: Knowledge Sources.
Memory and Retrieval
What is the difference between knowledge sources and memory collections?
Memory collections are where files are stored and indexed. Knowledge source configuration controls how an agent retrieves from that memory at runtime.
Why is my agent ignoring uploaded documents?
Usually one of these:
- Files are not fully indexed yet
- The collection is not attached to the agent
- Prompt does not ask for source-grounded output
Read: Memory System.
Can agents and users both create memory entries?
Yes. Users can upload directly, and agents can write to memory when write capability is configured.
What file formats are best for retrieval quality?
Use clean, well-structured source files (for example, Markdown, DOCX, and PDFs with selectable text). Keep documents focused by topic.
Chat and Collaboration
What are the three chat types in Arkios?
- Home chat for quick AI interactions
- Agent chat for purpose-configured workflows
- Group chat for team collaboration with AI and non-AI modes
Read: Chat Interface.
When should I use group chat instead of agent chat?
Use group chat when multiple teammates need to collaborate in one thread and optionally invoke AI in context.
Can I switch between AI and non-AI collaboration in group chat?
Yes. Group chat supports both direct member conversation and agent-assisted interaction in the same workspace context.
Artifacts and Reuse
I got a good response. What is the right way to keep it?
Use Save Response and store it as an artifact so it can be reviewed, shared, and reused.
Read: Artifacts.
Are artifacts only for final deliverables?
No. You can save drafts, share for review, then iterate until ready.
Teams, Access, and Governance
My teammate cannot access an agent I can see. Why?
This is usually a visibility or role permission mismatch.
Read: Managing Teams and Roles & Permissions.
How should we roll out safely to a larger team?
Use a staged rollout:
- Validate with a small pilot group
- Restrict tool access initially
- Expand visibility after stable results
Troubleshooting
The output quality dropped after I changed prompt or tools. What should I do?
Roll back one change at a time and retest with the same prompt set. Most regressions come from instruction drift, irrelevant tools, or mismatched memory sources.
Where should I go if this FAQ does not cover my issue?
Last updated: April 3, 2026