Silktide Agent
Silktide Agent is the conversational assistant available throughout Silktide. It combines product knowledge with permitted account data and supported actions, so it can help you investigate results and complete work without leaving your current screen.
The Agent can make mistakes. Verify important answers before you act on them.
When you start an AI report, the Agent sidebar shows that run while you stay on the screen you started from. The report opens itself when it is ready to read.
From the Agent sidebar you can also talk to a person on the Silktide team, book a call with Customer Success, or open the public product roadmap.
Practical examples
Set up and test websites
Test 100 pages of hsbc.com.
I'll create hsbc.com, configure a 100-page test, and start it. If anything blocks setup (login walls, robots rules, or missing permissions), I'll ask before continuing.
Retest all of my staging websites.
I found 4 staging websites matching that description. I'll retest all of them with their current settings and tell you when each run finishes.
Can you spell check oliveremberton.com?
I'll run a one-off spelling check on oliveremberton.com without adding it as a monitored website. Results will list any misspellings I find on that page.
Understand analytics and account data
Where are our visitors coming from?
Here are the top countries for the last 30 days from Analytics:
| Country | Sessions | Share |
|---|---|---|
| United States | 42,180 | 38% |
| United Kingdom | 21,450 | 19% |
| Germany | 9,820 | 9% |
| Canada | 7,110 | 6% |
| Australia | 5,940 | 5% |
Most traffic is from English-speaking markets. I can break this down by city, device, or landing page if that helps.
How many administrators do we have in the US?
You have 7 administrators with a US location. I can list them by name, email, and last sign-in if you want.
Investigate pages and issues
What is wrong with the alt text on our homepage?
I checked the homepage images. Two need better :
- Hero banner: currently empty. Suggest: “Team collaborating in a bright office.”
- Product screenshot: currently “image1.png”. Suggest: “Silktide dashboard showing accessibility scores.”
The logo and decorative icons look fine. I can draft replacements for the two issues if you want.
How did you find this page?
From the Inspector, I can see this page was discovered via:
- 3 incoming links from other pages on the same website
- Inclusion in your XML sitemap
So it is linked from the site and declared in the sitemap, not only reached by a direct URL.
Organise your account
Set up these users and their permissions for me.
I can create the users and roles from your spreadsheet. Attach the file (or paste the rows), and I'll confirm anything unclear before making changes.
Add appropriate tags to all my websites.
I'll review your websites and propose a tagging scheme (for example by brand, region, or environment). You'll approve the plan before I apply any tags.
Create a custom check that finds 'Oliver Emberton' on a page, and test it on oliveremberton.com.
I'll draft a custom check that looks for the text Oliver Emberton, then run it once against oliveremberton.com so you can see the result before you keep it.
Ask what you can do about a Presence finding
On a Presence fact check or an assistant answer, What can I do? opens the Agent with that finding already attached. The Agent looks up the fact or response, explains why the platform said what it said, and proposes work. You stay on the screen you were reading.
Turn a conversation into work
Create a support ticket for that.
I'll turn this discussion into a proposed Silktide task. Open it from the link to review or change it. If you have a task tracker like Linear connected, accepting the proposal can also open the work there.
Context and tools
The Agent may receive context about the screen you are viewing. It can use permitted account data and return links that open the exact result it found.
The Agent does not have unrestricted access to the product or the internet. It follows the same permissions as the rest of Silktide.
Available actions depend on your account features, permissions, connected integrations, and the capabilities currently supported by the Agent. It may ask you to clarify scope or approve changes before continuing.
Conversations and attachments
Conversations have durable message history so they can continue across runs and model choices. You can attach relevant files when a task needs them. Only attach information that is appropriate to process with AI.
Thinking effort and model
Beside the message box you can choose how hard the Agent thinks, with a slider from the lowest effort the model offers to the highest. Lower effort answers quickly; higher effort reasons for longer and is better on genuinely difficult problems. The levels on offer come from the model itself, so their names vary.
If your account has enabled more than one model, you can also change which one answers from the Model row in the Agent's message box. Most people never need to: an administrator sets the account default. See AI models and providers.
Each reply shows the model and effort that produced it, so you can always tell what answered.
When to verify independently
Treat legal interpretations, policy claims, accessibility certification, destructive changes, and high-impact recommendations as requiring human review. The Agent is a powerful working assistant, not an independent authority.