Frequently asked questions

The most common questions about evaluating Syfo, getting started, everyday use, billing, permissions, and extending tools, each with a short answer and a link to the details.

Syfo and your existing tools

We already use Feishu / WeCom / Slack. Why do we need Syfo?

Syfo doesn’t ask you to move all your communication over. It’s for work where “discussion turns into action, AI takes part in the execution, and the team needs to see the process and results”: the Agent is right there in the channel, sees the context, takes tasks, updates status, and delivers results. When needed, Agents can also read and write Feishu docs, multi-dimensional tables, and more through command-line tools or integrations, then write results back to your existing systems. See What is Syfo.

We have lots of internal systems. How does Syfo connect to them?

You don’t need to move your business systems into Syfo. Start with clear tool entry points: let Agents read the information they need and perform stable actions through APIs, command-line tools, MCP, or skills, then widen the scope step by step according to permissions. See MCP connections, Integrations, and Skills; if you need external systems to push events into a channel, see Webhook.

Can Agents only work by clicking through web pages in a browser?

No. Browser automation is a last resort. In production, prefer APIs, command-line tools, MCP, or skills. They can be tested and audited, and a page redesign won’t break your workflow.

Do you support private deployment?

We can discuss it. Contact sales for an assessment. See Security & permissions.

Creating and configuring Agents

Where should I start when creating an Agent?

Pick a real scenario first, then work backward to the channels, Agents, and tools it needs. Don’t start from a blank prompt. For the steps, see Create your first Agent; for planning, see Team playbook.

When should I create a new Agent, and when should I reuse one?

If it’s the same kind of task with a different subject, reuse. If the responsibilities, permissions, external systems, or sensitive data scope are clearly different, create a new one. Start with one Agent and split only when responsibilities really diverge. See Channels & collaboration.

What’s the difference between a local computer and Syfo Cloud?

On My computer, the Agent can use the repositories, tools, and sign-ins already on your machine, but it pauses when the computer is off or offline. Syfo Cloud is always online and billed in organization credits; the Owner and admins can create Cloud Agents directly, while regular members need to submit a request. See Computers and run locations.

My computer already has Claude Code / Codex installed, with skills and MCP set up. How do I bring it into Syfo?

First connect this computer to Syfo (“Computers → Add computer”), then create an Agent on this computer. It will use the configuration you already have locally. Agents on Syfo Cloud don’t inherit your computer’s configuration; you need to install things for them separately. You can see installed skills in the “Skills” tab on the Agent’s detail page, and MCP in the “MCP” tab. See Computers and run locations, Skills, and MCP connections.

Billing

Does it cost anything to run an Agent on my own computer?

If the Agent uses the account or key from the runtime on your computer, Syfo doesn’t charge credits. Model costs come out of the organization balance only when “Model payment” is set to “Organization credits”. See Usage & billing.

What do credits pay for?

Mainly four things: model calls (LLM) on Syfo Cloud and those paid with organization credits, Powerpack, hosted compute, and website hosting. See Usage & billing.

What happens when credits run out?

Agents that need credits pause, Powerpack calls don’t run, and websites stop serving after a grace period. Everything resumes automatically after you top up. When the balance is running low, the Owner and admins get a warning in advance. See Usage & billing.

Why can’t I see “Usage” and “Subscription & Credits”?

These two are only available to the organization Owner and admins. Contact them.

Channels and collaboration

How many Agents should a channel have?

Start small. For a standing project, first get the workflow running with “multiple people + 1 resident Agent”, then add more Agents by responsibility when you need more roles. See Channels & collaboration.

Won’t multiple Agents get chaotic?

They can, which is why you shouldn’t pull many Agents into one channel from the start. First decide who breaks down tasks, who executes, who checks, and which actions need a person to confirm. See Channels & collaboration and Message delivery.

I assigned a scheduled task to one Agent. Why did every Agent in the channel get it?

When a scheduled task lives in a thread or task, only the target Agent is notified. When it’s posted in the channel’s main message stream, Agents in the channel are notified according to the channel’s delivery setting. When you create a scheduled task, name the Agent that should run it, and keep follow-up discussion in the same task or thread. See Scheduled tasks and Message delivery.

Why does the Agent file my follow-up questions under the same task?

If you keep asking in the same thread or task, the Agent assumes you’re adding to the same piece of work. To start a new task, go back to the channel’s main message stream and @ it again, or say outright “this is a new task”. See Tasks.

What’s the difference between files and artifacts?

Files are working material, such as screenshots, logs, and reference docs. Artifacts are final results, such as reports, deployment URLs, and final PDFs. See Files & artifacts.

How do I adjust notifications so I’m not interrupted?

Each channel can be set to “All messages”, “@mentions and followed threads”, or “Mute”. You can also turn on “Mute this organization” in “Settings → Notifications”. See Notifications.

Agent behavior and tuning

How do I get an Agent to start working?

@ it in the channel and state the goal, inputs, output format, and acceptance criteria:

@research-assistant Collect the product issues customers raised in this channel this week, sort them into P1/P2/P3, and suggest actions.

For work that needs follow-up, the Agent turns it into a task. See Tasks.

The Agent keeps showing as working and doesn’t reply. What should I do?

Don’t keep @mentioning it. Go to the “Activity” tab on the Agent’s detail page to see what it’s doing. The task may simply be long, messages may be queued, or the computer it runs on may be offline or its runtime may have a problem. If there’s no progress for a long time, ask the Agent’s owner or an organization admin to restart the Agent, or check the computer it runs on. See Computers and run locations.

Why does the Agent sometimes take a minute or two?

An Agent is a long-running session that may be working across several channels and tasks at once. While a previous tool call is still running, later messages wait in a queue. Splitting channels by business area, keeping each Agent focused on one or two modules, and resetting the session when it’s idle all help it respond faster. See Best practices.

Why does the Agent keep stopping to ask me to confirm?

Usually its working boundaries weren’t agreed clearly, or the session is too long or too busy. Write “which actions it can move forward on its own and which must be confirmed” into its long-term memory. See Best practices.

The Agent creates a branch and reads the whole repository just to change something small. Is that normal?

That usually means its working habits haven’t been tuned yet. Tell it: make small changes directly on the current branch, use a feature branch for medium or larger changes, use git worktree when several Agents are changing the same repository, and find the relevant files first before widening the scope. See Best practices.

What if the Agent gets something wrong?

Point out the problem directly in the thread and give it a new constraint, for example “Only keep issues customers explicitly mentioned. Don’t add your own guesses.” For high-risk actions, have it present the details for confirmation first, and run them only after a person approves. See Approvals.

How do I pause, remove, or reset an Agent?

After you remove an Agent from a channel, it no longer receives that channel’s messages. Resetting or restarting is done by the Agent’s owner or an organization admin. See Permissions.

Development feels like a black box. How do I know what changed, and how do I roll back?

Keep the Syfo conversation and the Agent’s “Activity” tab on one screen, and your editor or terminal with git diff and the commit history on another. Leave versions and rollbacks to git, and have the Agent explain which files it changed, why, and how to verify.

Permissions and security

Can Agents see my private data?

Agents can read public channels. For private channels and direct messages, they must be members to see anything, and they can’t see direct messages between other members. External systems need separate authorization. See Security & permissions.

Why can’t I send a direct message to an Agent?

By default, only the Agent’s owner can DM it. The owner can turn on “Allow member DMs” under “Permissions” → “DM access” on the Agent’s detail page. You can click “Request DM access” to contact the owner. See Security & permissions.

Why can’t I see a certain channel?

Only invited members can see a private channel. Ask a member of the channel, or the channel owner, to add you. See Permissions.

Why can’t I create a Syfo Cloud Agent?

The Owner and admins can create one directly; regular members need to fill in a request and pick an Owner or admin to approve it. You also can’t create one when the organization has no credits available. See Computers and run locations.

Where do credentials go when an Agent uses an external system?

In “Settings → Vault”. Then grant them to Agents one by one. Don’t post them directly in a channel. See Vault.

What channel do model calls go through?

Through the model vendor’s official channel or an authorized channel from a major provider. Request content is sent to the corresponding model provider. See Security & permissions.

Which actions must a person confirm?

External sends, payments, production releases, data deletion, permission changes, and anything else that’s hard to roll back. See Approvals.

Tools and extensions

Can an Agent run tasks automatically on a schedule?

Yes. Just tell the Agent the time and what to do:

@research-assistant Every day at 9 a.m., summarize yesterday's discussion in #product-team into a brief and post it here.

The Agent needs to be online at the scheduled time. For long-term, reliable runs, use Syfo Cloud. See Scheduled tasks.

Can Agents generate images and videos?

Yes. Once your organization enables Powerpack and installs it on the computer the Agent runs on, the Agent can generate and edit images, videos, voiceovers, and background music, and look up e-commerce and social media data. Usage is billed in organization credits.

Can an Agent publish a website for me?

Yes. The Agent handles the build and deployment, and before going live it sends an approval card for the website Owner or an organization admin to approve. See Websites.