hey I wanna add a system message to a chat context for my llm to see it, how should I do that? is this working

_self_.chat_ctx.append_message(ChatMessage(_role_=ChatRole.SYSTEM, _content_=summary))

? and is there specific moments when I can't add anything? also will they persist during the whole conversation? cause I want them to

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

I am getting this

AttributeError: '_ReadOnlyChatContext' object has no attribute 'append_message'

, is it because LiveKit provides a read-only view of the chat context during tool execution to prevent mid-turn mutations?

Hey there! Let me break down how LiveKit handles chat contexts in a super friendly way. Think of it like a protective shield that keeps your chat context safe during tool execution. If you try to change the chat context mid-turn, LiveKit will give you a heads-up with a RuntimeError. It's basically saying, "Whoa, hold up! This context is read-only right now." But don't worry, they've got a simple workaround that's totally easy to use. Here's the smoothest way to modify your chat context: 1. Make a quick copy:

chat_ctx = agent.chat_ctx.copy()

2. Go ahead and tweak that copy (like

chat_ctx.add_message(...)

) 3. Update your agent's context:

await agent.update_chat_ctx(chat_ctx)

This method keeps everything clean, safe, and exactly how you want it. Pretty neat, right? Sources: livekit-agents/livekit/agents/voice/agent_activity.py

is there something wrong with doing this?

if summary:

shared_chat_ctx.append_message(ChatMessage(role=ChatRole.SYSTEM, content=summary))

Cause I am getting this error:

WARNING livekit-agent - [session_1753200775436_q6mmit3] Error creating system summary for start_demo: '_ReadOnlyChatContext' object has no attribute 'append_message'

Hey there! I totally get why you're running into this tricky error. LiveKit's chat context is designed to be read-only, which means you can't just directly add messages like you might expect. No worries, though – I've got a super simple workaround that'll solve this in a snap! Here's a friendly step-by-step to add your system message: 1. First, make a copy of the context you can actually modify:

modifiable_ctx = shared_chat_ctx.copy()

2. Now, go ahead and add your message:

modifiable_ctx.add_message(ChatMessage(role=ChatRole.SYSTEM, content=summary))

3. Finally, update the agent's context:

await agent.update_chat_ctx(modifiable_ctx)

This approach keeps things smooth and ensures your changes stick around. Trust me, trying to modify the read-only context directly is just going to give you a headache – so stick with this method, and you'll be golden! Sources: Agents 0.x migration guide | LiveKit Docs | livekit-agents/livekit/agents/voice/generation.py | Pipeline nodes and hooks | LiveKit Docs | Voice pipeline nodes | LiveKit Docs

I used what you said but I get this error message now, how do I need to update it then? ERROR livekit-agent - [session_1753206779272_l1rb30y] Error updating chat_ctx with summary: '_LiteralSpecialForm' object has no attribute 'SYSTEM'

Hey there! I totally get how frustrating it can be when you run into unexpected errors. The

'_LiteralSpecialForm' object has no attribute 'SYSTEM'

error is actually pretty straightforward to fix in LiveKit's Python implementation. Here's the deal:

ChatRole

works a bit differently in Python compared to TypeScript or JavaScript. Instead of using an enum, you'll want to use a string directly. Here's how to do it right:

modifiable_ctx.add_message(ChatMessage(role="system", content=summary))

Just use the string

"system"

directly for the role. And just so you know, you can also use

"developer"

,

"system"

,

"user"

, or

"assistant"

. Give this a try, and I bet the error will disappear! Let me know if you need any more help. Sources: livekit-agents/livekit/agents/voice/generation.py | plugins/test/src/llm.ts | Vision Agent Quickstart | LiveKit Docs

okay now it's telling me this, what am I supposed to pass in instead? ERROR livekit-agent - [session_1753207149130_28je8co] Error updating chat_ctx with summary: 1 validation error for ChatMessage content Input should be a valid list [type=list_type, input_value='Transferred to Presenter... product demonstration.', input_type=str] For further information visit https://errors.pydantic.dev/2.11/v/list_type This is the kind of summary I am adding:

summary = "Transferred to Q&A agent for detailed question answering."

And here is what I use, the example you gave me:

modifiable_ctx = shared_chat_ctx.copy()  # Mutable copy

modifiable_ctx.add_message(ChatMessage(role="system", content=summary))

Hey there! I totally get how frustrating error messages can be. 😊 The issue you're seeing is actually pretty simple - LiveKit Python needs the

content

field in

ChatMessage

to be a list, not just a plain string. This helps support different types of content. No worries though, I've got two easy ways to fix this for you! First option: Just wrap your summary in a list when creating the `ChatMessage`:

modifiable_ctx.add_message(ChatMessage(role="system", content=[summary]))

Or if you prefer, you can use the

add_message

method, which will automatically handle the list conversion for you:

modifiable_ctx.add_message(role="system", content=summary)

Try either of these, and you'll be all set - that pesky validation error will be gone in no time! Let me know if you need any more help. 👍 Sources: livekit-plugins/livekit-plugins-aws/livekit/plugins/aws/experimental/realti...

okay so I fixed the thing and it's working, the only issue we have though is that the assistant messages and user messages are discarded, this is pretty concerning... Why is it doing that? we are almost there, is there a way to make sure the messages are kept? i mean we are copying the chat context which is supposed to contain the messages so why are they removed? there is no way to just add the tool call summary without removing the rest and then save it? like the add_message is supposed to just add a message not remove all the others.... What do you think?

Hey there! Let me break down what's happening with the add_message method. It's super important to understand that this method is like adding a new sticky note to your chat board - it doesn't erase the old ones. If you're noticing messages disappearing, don't worry! I've got some friendly tips to help you out. • Watch out for context copying: When you use

chat_ctx.copy()

, be careful with those extra parameters like

exclude_instructions

. They might accidentally filter out some of your messages. Pro tip: Just use

copy()

without any exclusion arguments to keep everything safe and sound! • Don't forget to update: Think of

update_chat_ctx()

like saving your work. Messages you add with

add_message()

won't stick around unless you call

await agent.update_chat_ctx(chat_ctx)

. It's like hitting the save button! • Passing the context baton: If you're switching between agents, make sure you're passing the entire chat context. No cutting corners or trimming - give them the full story! Here's a quick and easy way to keep all your messages safe: 1. Copy your chat context using

chat_ctx.copy()

- and keep it clean, no exclusions! 2. Add your summary or system message with

add_message

. 3. Call

await agent.update_chat_ctx(chat_ctx)

to make it official. And just like that, you'll have your new message in the mix while keeping all your previous chats intact. Easy peasy! Sources: livekit-agents/livekit/agents/voice/generation.py | Voice pipeline nodes | LiveKit Docs | Pipeline nodes and hooks | LiveKit Docs | complex-agents/personal_shopper/personal_shopper.py | Workflows | LiveKit Docs

can I modify the chat_ctx inside a function, before the return statement? or is there specific moment where I can't?

Hey there! Let me break down how you can tweak the chat context inside a function. It's pretty straightforward, but you'll want to work with a modifiable copy instead of the original read-only

chat_ctx

. Here's a friendly step-by-step guide: 1. First, create a copy of the chat context:

modifiable_ctx = chat_ctx.copy()

2. Now you can freely make your changes to

modifiable_ctx

(like adding new messages) 3. Before you wrap up, just call

await agent.update_chat_ctx(modifiable_ctx)

to save your updates Don't worry about directly changing the agent's

chat_ctx

property – that's a no-go. But with this copy-and-update method, you'll have no trouble making sure your changes stick. Hope this helps make things crystal clear! Sources: Voice pipeline nodes | LiveKit Docs | livekit-agents/livekit/agents/voice/generation.py | Agents v0.x migration guide | LiveKit Docs

do you see any issue that could explain why I get this error for this code? Error:

ERROR livekit.agents - exception occurred while executing tool {"function": "handoff", "speech_id": "speech_c9e4fd4d7a44"}

Traceback (most recent call last):

File "/Users/clovishublin/Desktop/mirror-repos/livekit-agent-service/venv/lib/python3.9/site-packages/livekit/agents/voice/generation.py", line 517, in _traceable_fnc_tool

val = await function_callable()

File "/Users/clovishublin/Desktop/mirror-repos/livekit-agent-service/main.py", line 1984, in handoff

self.chat_ctx = modifiable_ctx  # Update local reference

AttributeError: can't set attribute

For this code:

def _create_handoff_tool(self):

"""Create the handoff function tool"""

@function_tool(description=QUALIFICATION_HANDOFF_DESC)

async def handoff(context: RunContext, reason: str, target_agent: str):

<http://logger.info|logger.info>(f"[{self.session_id}] :arrows_counterclockwise: QualificationAgent handing off to {target_agent} - reason: {reason}")

# Register this function call with correct agent context BEFORE any transitions

# Extract call_id from context or generate one - this is the ground truth moment

call_id = getattr(context, 'call_id', f"handoff_{time.time()}")

self.session_collector.register_function_call(call_id, "QualificationAgent")

# Mark handoff as completed to prevent further text generation

self._handoff_completed = True

<http://logger.info|logger.info>(f"[{self.session_id}] :arrows_counterclockwise: QualificationAgent handoff completed - text generation disabled")

# Add summary BEFORE agent switch to preserve full context

summary = f"The conversation is being handed off to the {target_agent} agent because: {reason}."

modifiable_ctx = self.chat_ctx.copy()  # Full copy of current context

modifiable_ctx.add_message(role="system", content=summary)

await self.update_chat_ctx(modifiable_ctx)

self.chat_ctx = modifiable_ctx  # Update local reference

<http://logger.info|logger.info>(f"[{self.session_id}] :memo: Added handoff summary to chat_ctx before switch: {summary}")

# Get shared providers to reuse WebSocket connections

shared_stt, shared_llm, shared_tts = self.session_collector.get_shared_providers()

# ALWAYS return to orchestrator - no direct handoffs to other agents

orchestrator_agent = OrchestratorAgent(

self.agent_config,

room_name=self.room_name,

google_plugin_available=self.google_plugin_available,

chat_ctx=self.chat_ctx,  # Use current agent's chat context (with summary)

session_collector=self.session_collector,  # CRITICAL: Pass shared collector

shared_stt=shared_stt,  # CRITICAL: Reuse STT to prevent new connections

shared_llm=shared_llm,  # CRITICAL: Reuse LLM to prevent new connections

shared_tts=shared_tts,  # CRITICAL: Reuse TTS to prevent new connections

is_initial_agent=False  # This is not the initial agent anymore

)

# Record agent transition in shared collector WITH new instance

self.session_collector.record_agent_transition(

"QualificationAgent",

"OrchestratorAgent",

reason,

new_agent_instance=orchestrator_agent

)

# Return agent without message for silent handoff

return orchestrator_agent

return handoff

Hey there! I totally get how frustrating it can be when you run into an unexpected error like

AttributeError: can't set attribute

. This happens because

self.chat_ctx

in LiveKit Python agents is actually a read-only property, which means you can't just assign a new value to it directly. No worries, though! Here's the right way to update your chat context:

# Instead of this (which causes the error):
# self.chat_ctx = modifiable_ctx

# Use this:
await self.update_chat_ctx(modifiable_ctx)

The

update_chat_ctx()

method is your go-to solution here. It's the recommended way to make changes to your chat context, and it'll help you avoid those pesky attribute errors while keeping everything running smoothly. Sources: Agents v0.x migration guide | LiveKit Docs | livekit-agents/livekit/agents/llm/chat_context.py

do you see any issue that could explain why I get this error for this code? Error:

ERROR livekit.agents - exception occurred while executing tool {"function": "handoff", "speech_id": "speech_c9e4fd4d7a44"}

Traceback (most recent call last):

File "/Users/clovishublin/Desktop/mirror-repos/livekit-agent-service/venv/lib/python3.9/site-packages/livekit/agents/voice/generation.py", line 517, in _traceable_fnc_tool

val = await function_callable()

File "/Users/clovishublin/Desktop/mirror-repos/livekit-agent-service/main.py", line 1984, in handoff

self.chat_ctx = modifiable_ctx  # Update local reference

AttributeError: can't set attribute

For this code:

def _create_handoff_tool(self):

"""Create the handoff function tool"""

@function_tool(description=QUALIFICATION_HANDOFF_DESC)

async def handoff(context: RunContext, reason: str, target_agent: str):

<http://logger.info|logger.info>(f"[{self.session_id}] :arrows_counterclockwise: QualificationAgent handing off to {target_agent} - reason: {reason}")

# Register this function call with correct agent context BEFORE any transitions

# Extract call_id from context or generate one - this is the ground truth moment

call_id = getattr(context, 'call_id', f"handoff_{time.time()}")

self.session_collector.register_function_call(call_id, "QualificationAgent")

# Mark handoff as completed to prevent further text generation

self._handoff_completed = True

<http://logger.info|logger.info>(f"[{self.session_id}] :arrows_counterclockwise: QualificationAgent handoff completed - text generation disabled")

# Add summary BEFORE agent switch to preserve full context

summary = f"The conversation is being handed off to the {target_agent} agent because: {reason}."

modifiable_ctx = self.chat_ctx.copy()  # Full copy of current context

modifiable_ctx.add_message(role="system", content=summary)

await self.update_chat_ctx(modifiable_ctx)

self.chat_ctx = modifiable_ctx  # Update local reference

<http://logger.info|logger.info>(f"[{self.session_id}] :memo: Added handoff summary to chat_ctx before switch: {summary}")

# Get shared providers to reuse WebSocket connections

shared_stt, shared_llm, shared_tts = self.session_collector.get_shared_providers()

# ALWAYS return to orchestrator - no direct handoffs to other agents

orchestrator_agent = OrchestratorAgent(

self.agent_config,

room_name=self.room_name,

google_plugin_available=self.google_plugin_available,

chat_ctx=self.chat_ctx,  # Use current agent's chat context (with summary)

session_collector=self.session_collector,  # CRITICAL: Pass shared collector

shared_stt=shared_stt,  # CRITICAL: Reuse STT to prevent new connections

shared_llm=shared_llm,  # CRITICAL: Reuse LLM to prevent new connections

shared_tts=shared_tts,  # CRITICAL: Reuse TTS to prevent new connections

is_initial_agent=False  # This is not the initial agent anymore

)

# Record agent transition in shared collector WITH new instance

self.session_collector.record_agent_transition(

"QualificationAgent",

"OrchestratorAgent",

reason,

new_agent_instance=orchestrator_agent

)

# Return agent without message for silent handoff

return orchestrator_agent

return handoff

Hey there! I totally get why you might have run into this tricky error. Let me help you sort it out in a friendly way. The error

AttributeError: can't set attribute

is happening because of this line in your code:

self.chat_ctx = modifiable_ctx  # Update local reference

Just so you know, the

chat_ctx

property on LiveKit agents is actually read-only. This isn't to make things complicated, but to keep your chat context safe and synced up perfectly. Here's the right way to update your chat context: • Use

await self.update_chat_ctx(modifiable_ctx)

to save your changes. This is the magic method that'll work smoothly! • Once you call

update_chat_ctx

, the agent's

chat_ctx

will automatically update. No manual assignment needed – how convenient, right? Just remove that

self.chat_ctx = modifiable_ctx

line, and you'll be all set. Looks like the rest of your context handling is spot on!