AI and Rules - Recent Threads

RE: AI Skill

khushimurkya — Thu, 09 Jul 2026 12:25:15 GMT

Is this issue solved or still open?

AI Skill

kunalj445796 — Tue, 16 Jun 2026 06:22:04 GMT

Hello All,

I am using Version 26.3 and while additing new AI Skill, i am getting blank screen everytime. I cleared cache, restarted browser and system but with no luck. Even tried with other bowsers also. Within my organization, the same is working fine for few but there are few who are facing issue similar to what i have been facing. Please suggest.

Br,

Kunal

Chat Agent – Tool Execution Failure ("Downstream tool request failed")

Kaviarasu Mohan — Thu, 09 Jul 2026 11:00:15 GMT

Hi Team,

Our chat agent is intermittently failing when attempting to execute tool calls. End users are receiving one of the following error messages during these interactions:

Tool execution failed: Downstream tool request failed
I'm currently experiencing a system issue retrieving the data.

RE: Chat Agent – Tool Execution Failure ("Downstream tool request failed")

Kaviarasu Mohan — Thu, 09 Jul 2026 12:22:58 GMT

Hi kapilsharma ,

Thank you for your reply.

I am new to Appian and Chat Agents. Could you please help me understand where I can check and find the idle timeout and read timeout settings for the downstream service?

Thank you for your assistance.

RE: Chat Agent – Tool Execution Failure ("Downstream tool request failed")

kapilsharma — Thu, 09 Jul 2026 11:52:18 GMT

Hi Kaviarasu Mohan ,

Server-Sent Events (SSE) require a long-lived, persistent HTTP connection to stream the LLM's response back to the user. If your downstream tool takes 30 to 60 seconds to execute, the AI pauses streaming.
I would suggest check the idle timeout and read timeout settings of your downstream service.

Moreover, the Appian chat component is waiting for the AI, and the AI is waiting for your tool to finish its work. If the tool hangs or takes too long, the underlying TCP socket simply gives up and drops.

Can you try increasing these timeouts to accommodate the maximum duration your longest tool takes to run, plus the standard LLM processing time.

RE: Chat Agent – Tool Execution Failure ("Downstream tool request failed")

Kaviarasu Mohan — Thu, 09 Jul 2026 11:41:46 GMT

Hi Kapil,

When i checked the log, this is the error message we are getting.

OkHttp ai-platform.ai-platform.svc.cluster.local/...] ERROR com.appiancorp.lcpapi.streaming.StepStreamingEventSourceListener - Upstream SSE stream failed for run
java.net.SocketException: Socket closed
at java.base/sun.nio.ch.NioSocketImpl.ensureOpenAndConnected(NioSocketImpl.java:165)
at java.base/sun.nio.ch.NioSocketImpl.beginRead(NioSocketImpl.java:236)
at java.base/sun.nio.ch.NioSocketImpl.implRead(NioSocketImpl.java:304)
at java.base/sun.nio.ch.NioSocketImpl.read(NioSocketImpl.java:355)
at java.base/sun.nio.ch.NioSocketImpl$1.read(NioSocketImpl.java:808)
at java.base/java.net.Socket$SocketInputStream.read(Socket.java:966)
at okio.InputStreamSource.read(JvmOkio.kt:93)
at okio.AsyncTimeout$source$1.read(AsyncTimeout.kt:128)
at okio.RealBufferedSource.request(RealBufferedSource.kt:209)
at okio.RealBufferedSource.require(RealBufferedSource.kt:202)
at okio.RealBufferedSource.readHexadecimalUnsignedLong(RealBufferedSource.kt:384)
at okhttp3.internal.http1.Http1ExchangeCodec$ChunkedSource.readChunkSize(Http1ExchangeCodec.kt:437)
at okhttp3.internal.http1.Http1ExchangeCodec$ChunkedSource.read(Http1ExchangeCodec.kt:416)
at okhttp3.internal.connection.Exchange$ResponseBodySource.read(Exchange.kt:281)
at okio.RealBufferedSource.select(RealBufferedSource.kt:232)
at okhttp3.internal.sse.ServerSentEventReader.processNextEvent(ServerSentEventReader.kt:50)
at okhttp3.internal.sse.RealEventSource.processResponse(RealEventSource.kt:75)
at okhttp3.internal.sse.RealEventSource.onResponse(RealEventSource.kt:46)
at okhttp3.internal.connection.RealCall$AsyncCall.run(RealCall.kt:519)
at java.base/java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1136)
at java.base/java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:635)
at java.base/java.lang.Thread.run(Thread.java:840)

RE: Chat Agent – Tool Execution Failure ("Downstream tool request failed")

kapilsharma — Thu, 09 Jul 2026 11:12:52 GMT

Review Appian's integration logs (tomcat-stdOut.log or the specific integration error logs in the Monitoring view) for HTTP 429 Too Many Requests errors that correlate with the exact timestamps.

RE: AI Agent Performance

Mallepu Prashanth — Thu, 09 Jul 2026 10:45:19 GMT

Thanks, adding to that I'm getting this error: Agent 250512b0-a322-45b6-85c5-525b87755a8c could not be found. When I try to include it in the Process model. I have checked UUID is correct and the environment has active and enabled.

AI Agent Performance

Mallepu Prashanth — Thu, 09 Jul 2026 10:16:13 GMT

How accurate AI Agent works same task with different data. Will it hallucinate or imagine things and provide output? Does it reconcile based on the repeated tasks or does it take every new data as new task?

RE: AI Agent Performance

Shubham Aware — Thu, 09 Jul 2026 10:31:03 GMT

AI Agents are non‑deterministic and stateless: each Execute AI Agent run starts fresh from the current prompt, inputs, and tools, can produce slightly different outputs, and does not remember or learn from previous runs. Reliability comes from how you constrain it with rules, tools, and validations in the process model, not from the agent itself.

https://docs.appian.com/suite/help/26.6/agent-studio-best-practices.html

RE: AI Agent Performance

kapilsharma — Thu, 09 Jul 2026 10:25:41 GMT

Hallucination Risks

Because Appian AI Agents are powered by underlying generative LLMs, they function as statistical prediction engines, not fact-checkers. If the input data you provide is ambiguous, missing, or unstructured, the agent will imagine plausible-sounding outputs to satisfy your prompt rather than failing gracefully. While Appian provides platform-level AI Guardrails, these are designed to prevent harmful content and enforce environment policies, not to guarantee the factual validity of your business data.

Task Reconciliation vs. New Tasks

I disagree with the premise that the agent reconciles based on repeated tasks because Appian's AI architecture is inherently stateless per execution. Here's what I'd do instead: Architect your process models to treat every single data injection as a brand new, completely isolated task.

If you need the agent to base its current decisions on past outcomes, you must provide it with explicit Appian Tools (such as a specific record query tool) that force it to read historical Appian Records before taking action. The risk in your approach is expecting the agent to organically learn or self-correct over time, which will inevitably lead to compounded errors and silent logic failures in your production workflows.

Getting an error in process model: Agent 250512b0-a322-45b6-85c5-525b87755a8c could not be found. AI Agent is not working in Process Model.

Mallepu Prashanth — Thu, 09 Jul 2026 10:03:56 GMT

When I test AI Agent it is working as expected and giving result. But when I incorporated it within Process model, it is giving error: Agent 250512b0-a322-45b6-85c5-525b87755a8c could not be found. So, I created another AI agent and getting same error stating Agent could not found. I have double checked the UUID of the AI Agent it is correct. What could be the reason?

In Admin console for our environment AI services are enabled and active.

AI Doc Center vs AI Skills

kl0001 — Thu, 09 Jul 2026 01:28:25 GMT

My use case is that I want to be classify documents into one of several categories, then depending on the category, extract certain fields from it. My question is for this kind of use case, what exactly is the difference between using AI Skills to do this versus using AI Doc Center? Looking around, I see some people saying AI Doc Center uses more tokens and that it can handle more complex cases, but I can't see anything in documentation about if that's actually true or not, and anything about when you'd want to use one or the other. I've been in loops reading the same page over and over again, so at this point I might have missed something, but if someone's able to help point me towards an answer, that would be most appreciated!

RE: AI Doc Center vs AI Skills

ujjwalr0002 — Thu, 09 Jul 2026 05:46:41 GMT

If your documents are already well-defined document types and your requirement is simply "if it's document type A, extract fields XYZ; if it's document type B, extract fields ABC," then I don't think I'd split the solution between AI Skills and AI Document Center. AI Document Center is capable of handling both the classification and the subsequent field extraction based on the identified document type.

The scenario where I'd lean towards AI Skills is when you only need classification (or some custom AI reasoning) and don't need the document-processing capabilities that AI Document Center provides.

[quote userid="70777" url="~/discussions/f/rules/40544/ai-doc-center-vs-ai-skills/155087"]Also, when you say 'review workflows or managing document extraction models' for AI Doc Center, what particular features are you referring to? What about workflows can you review in the AI Doc Center, and what can you manage regarding document extraction that you can't do in AI Skills?[/quote]

Lets say you are trying to extract some fields and confidentiality score is low in that case you can review that and proceed further that is what I mean by review workflow https://docs.appian.com/suite/help/26.6/aidc-4.2/reconcile-extraction.html

RE: AI Doc Center vs AI Skills

Shubham Aware — Thu, 09 Jul 2026 05:17:57 GMT

Use MyAppian -> Adoption -> AI Usage Metrics usage to see AI Action consumption by feature across environments.
Also from 26.6 in logs also :- https://docs.appian.com/suite/help/26.6/Logging.html#ai-usage-logs

RE: AI Doc Center vs AI Skills

kl0001 — Thu, 09 Jul 2026 04:40:18 GMT

Where exactly can you monitor tokens for Doc Center? In Doc Center, I can see Home, Classification, Extraction, Deployment, Documentation tabs on the left side, but nothing about token monitoring?

Also, where in the documentation does it say that AI Skills use no tokens? Want to just have a link as reference

RE: AI Doc Center vs AI Skills

kl0001 — Thu, 09 Jul 2026 04:37:03 GMT

Regarding extraction, I just need to be able to handle the following situation - if it's doc type A, then extract XYZ fields. if it's doc type B, then extract ABC fields. Sounds like then I'd probably want AI Skills to classify which document type it is, then use AI Doc Center to extract the values?

Also, when you say 'review workflows or managing document extraction models' for AI Doc Center, what particular features are you referring to? What about workflows can you review in the AI Doc Center, and what can you manage regarding document extraction that you can't do in AI Skills?

RE: AI Doc Center vs AI Skills

Shubham Aware — Thu, 09 Jul 2026 03:57:19 GMT

AI Skills are the runtime objects. You build them directly pick classification or extraction, train or prompt them, and call them from a process via smart services like Classify Document and Extract from Document.

Doc Center is not a separate engine. It's an application that generates and maintains those same AI skill objects and their process models for you, then layers on versioning, reconciliation, accuracy metrics, token monitoring, and deployment.

Both end up as AI skill objects at runtime. The choice is whether you build and manage the AI skill yourself, or let Doc Center generate it and give you versioning and accuracy metrics. Doc Center's models are gen-AI backed (tokens, better on varied layouts); an AI skill you build directly can be trained ML: no tokens, but needs consistent layouts and retraining.

RE: AI Doc Center vs AI Skills

ujjwalr0002 — Thu, 09 Jul 2026 03:10:56 GMT

From what I've learned so far, if your requirement is simply to classify documents and then route the process based on the classification, AI Skills would likely be the better approach. It gives you the flexibility to classify the document and drive the workflow accordingly.

If your use case goes beyond classification—for example, extracting different fields from multiple document types, handling confidence scores, review workflows, or managing document extraction models—then AI Document Center is probably the better fit since it's designed specifically for intelligent document processing.

Regarding token usage, I've seen community members mention that AI Skills may consume fewer tokens for simpler scenarios, but I haven't found any official Appian documentation that confirms this.

RE: AI Doc Center vs AI Skills

kl0001 — Thu, 09 Jul 2026 01:45:29 GMT

Also, I want to add that I want to classify/extract about 500 documents a day

RE: MCP Server endpoint URL in Appian 26.6

Sam Wahbeh — Wed, 08 Jul 2026 14:53:48 GMT

Thanks for pointing out the inconsistency. The admin console URL (<ENVIRONMENT_URL>/mcp)is the correct one. We will update the documentation to match.

MCP Server endpoint URL in Appian 26.6

Raúl Gómez Moya — Tue, 07 Jul 2026 19:24:15 GMT

Hello everyone,

I’m configuring the Appian MCP Server in Appian 26.6 and I noticed a difference between the endpoint URL shown in the Administration Console and the one described in the documentation.

In the Administration Console > MCP Server, the endpoint URL is displayed as:

<ENVIRONMENT_URL>/mcp

See Image 1.

However, the documentation for enabling the Appian MCP Server says that the endpoint URL follows this format:

<ENVIRONMENT_URL>/suite/mcp

docs.appian.com/.../enable-configure-mcp-server.html

See Image 2.

My question is:

Which endpoint should be used when configuring an external MCP client in Appian 26.6?

Should it be:

<ENVIRONMENT_URL>/mcp

or:

<ENVIRONMENT_URL>/suite/mcp

Has anyone configured this successfully and can confirm which URL is correct?

Thanks in advance.

Attachments

Image 1: Administration Console showing the MCP endpoint as /mcp.
Image 2: Documentation showing the MCP endpoint format as /suite/mcp.

How to Check AI Token Usage for AI Skills in Appian?

sureshs276515 — Tue, 07 Jul 2026 14:18:31 GMT

Hi Team,

We are evaluating AI Skills/Generative AI usage in Appian and would like to monitor token consumption for optimization and cost governance purposes.

I have a few questions:

Is there a way to view the input tokens, output tokens, and total tokens consumed by an AI Skill execution?
Does Appian expose token usage metrics at the AI Skill or process level?
What is the recommended approach to track and analyze token usage for prompt optimization?
Has anyone built a custom monitoring/dashboard solution to estimate token consumption for AI Skills and Agents?

thanks in advance.

RE: How to Check AI Token Usage for AI Skills in Appian?

Louis Prensky — Wed, 08 Jul 2026 14:39:07 GMT

Shubham Aware 's response is correct! I'd also like to add that, in 26.6, AI Actions are logged at the execution level for all AI features in the existing AI Usage Log, which is supported for log streaming. This feature is actually being introduced in this weekend's hotfix release, at which point the docs will be updated to reflect the new columns on the log file including AI Actions.

RE: MCP Server endpoint URL in Appian 26.6

khushimurkya — Wed, 08 Jul 2026 12:05:13 GMT

yes, i agree - I followed appian documentation but in admin console it's says different
https://docs.appian.com/suite/help/26.6/enable-configure-mcp-server.html in appian documentation it's says diferent.

Just to clarify, for me also it worked <ENVIRONMENT_URL>/mcp Hope it helps!