agentconfigupdate duplicates the same content as of system instructions in chat items

Thanks for reaching out, Test. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.

the content of chat item is same

Thanks — I appreciate the follow-up. I hear you: the chat item content is identical (the system instructions are being duplicated when

agentconfigupdate

runs). I want to help you get this fixed quickly. To diagnose whether the duplication is coming from the client UI, the server emitting duplicate messages, or the storage layer, please provide any of the following you can share: • Where you observe the duplicate content (choose one): web app UI, mobile app UI, server logs, database/chat history, or API responses. • One redacted example of the chat item that shows the duplicate (or a screenshot). Redact any sensitive fields. • The exact sequence to reproduce (step-by-step). For example: 1) open room, 2) run

agentconfigupdate

, 3) observe chat item. • Whether the duplicate appears immediately or after a refresh/reconnect. • Whether this happens every time or only sometimes. • The SDK or product version you’re running (e.g., Web SDK vX, Server SDK vY) and any recent changes to agent config or message-handling code. • If your system code triggers

agentconfigupdate

, please paste the short snippet that emits the event or describe how it’s published (redact secrets). Relevant details: message id generation, whether you attach the same

id

or timestamp, and whether you call the same publish function twice. • Any server logs around the time of the event (timestamps help). What I’ll do once you share those: • Confirm whether duplicate messages are being produced server-side or duplicated by the client. • Suggest targeted fixes (for example: dedup using message IDs/timestamps, ensure single emission of

agentconfigupdate

, or preventing system instruction echo on clients). • Provide step-by-step debugging commands or a code example for dedup logic if you want. If it’s easiest, paste one redacted chat item and the code that triggers

agentconfigupdate

and I’ll review it and point out likely causes and quick fixes. , with interactive elements

when agent_config_update type item is added to chat items

Thanks — I appreciate the clear update. I’m with you: the duplicate appears when an

agent_config_update

chat item is added, and we need to determine whether the duplication originates from the server, the client, or during storage/serialization. To troubleshoot quickly, please provide any of the following you can share (each will help me pinpoint the cause): • Where you see the duplicate (pick one): web UI, mobile UI, API response, server logs, or database/chat history. • One redacted example of the chat item as JSON (or a screenshot). Redact any secrets or personal data. I’m specifically looking for fields like

type

,

content

,

role

,

id

, and timestamps. • The exact steps to reproduce (short sequence). Example: 1) join room, 2) call publish/update, 3) chat item appears duplicated. • Whether duplication is immediate or only after refresh/reconnect. • Whether it happens every time or intermittently. • SDK/product and version you’re using (e.g., Web SDK vX, Server SDK vY). • If you emit the

agent_config_update

event from your code: paste the minimal snippet that publishes it (redact secrets). Important details: how you set `id`/timestamp, whether you call the publish function more than once, and whether you attach system instruction text when publishing. • Any server logs or timestamps around the event (two entries with same timestamp would indicate duplicate emission). What I’ll do after you share this: • Confirm whether duplicate entries are produced server-side or duplicated by the client renderer. • Recommend a targeted fix (for example: dedupe by message

id

, ensure single emission of the update, or avoid echoing system instructions to chat items). • Provide a short code example or steps to implement deduplication if needed. If it’s easiest, paste one redacted chat item JSON and the code that triggers

agent_config_update

and I’ll review it immediately and point to the most likely cause. , with interactive elements

are agent_config_update item types sent to llm or they are filtred?

Thanks — good question. Regarding

agent_config_update

items: they are included in the conversation history and sent to the LLM (they are not automatically filtered out). These items (which contain fields like

instructions

,

tools_added

, and

tools_removed

) are added when you update an agent’s tools or instructions so the LLM can see configuration changes. If you need to keep those updates out of a copied/serialized history, you can explicitly exclude them by using

exclude_config_update=True

when calling

chat_ctx.copy()

. If you want, tell me whether you’d like: • guidance on preventing duplicate UI/chat entries for these updates, or • a short code example showing how to copy history without config updates (using

exclude_config_update=True

). Sources: Function tool definition | LiveKit Documentation, with interactive elements

should we exclude them or include

Good question — whether to include

agent_config_update

items depends on what you need the LLM to know and the tradeoffs you’re willing to accept. Here’s a short guide to help you decide, plus the practical option to exclude them when desirable. When to include

agent_config_update

• You want the LLM to be aware of runtime changes to an agent’s instructions or toolset (important in multi-agent workflows or when agent behavior must adapt). • You need full auditability of configuration changes in the conversation history. • You accept the extra tokens and possible longer context so the model has that configuration context. When to exclude

agent_config_update

• You want to reduce token usage / keep prompts concise. • The config updates are only relevant for system bookkeeping and not helpful for the model’s next response. • You’re seeing UI duplication or noise in stored/serialized chats and prefer to keep the user-visible history clean. How to exclude them • When copying or serializing the chat context, use

exclude_config_update=True

with

chat_ctx.copy()

to omit these items from the copied history. Recommendation (best practice) • Default: include

agent_config_update

when configuration changes affect the agent’s behavior that the LLM should consider (e.g., tool additions/removals, instruction updates in multi-agent flows). • If the updates are only operational metadata, or you are trying to limit tokens or avoid UI duplication, exclude them when generating model input or building user-facing transcripts. If you like, I can: • Show a short code example of copying context without config updates. • Help decide for your specific flow if you paste a typical

AgentConfigUpdate

example or describe how you use those updates., with interactive elements