Skip to content
LOPS Docs

Gemini Live

v2.3.3Updated

The Gemini Live LOP provides real-time bidirectional voice conversation powered by Google’s Multimodal Live API. It streams microphone audio to Gemini and plays back the model’s spoken responses with ultra-low latency, while optionally sending video frames and text. Like the Agent LOP, it can discover and call tools from connected operators during a live conversation.

Key Features

Section titled “Key Features”
  • Real-time voice-to-voice conversation over WebSockets with automatic turn detection
  • Three turn management modes: Auto VAD, Push to Talk, and Hybrid
  • Video frame input (continuous streaming, manual pulse, or auto-send on connect)
  • Text injection during a live conversation
  • External tool orchestration via the same Tool sequence pattern used by the Agent LOP
  • Built-in Google Search grounding
  • Session history with save, load, and resumption support
  • Auto-reconnect on server errors with configurable retry

Requirements

Section titled “Requirements”
  • Google API Key: A Gemini API key with Live API access. Enter it on the Config page or store it in ChatTD’s Key Manager under the name gemini.
  • Python Dependency: The google-genai package. Pulse ‘Install/Update google-genai’ on the Config page to install it automatically.

Input/Output

Section titled “Input/Output”

Inputs

Section titled “Inputs”
  • Input 1 (Audio CHOP): Microphone audio fed to the operator via ReceiveAudioChunk. Typically a 16 kHz mono CHOP routed through an Audio Device In.

Outputs

Section titled “Outputs”
  • Output 1: Conversation table DAT (role, message, id, timestamp)
  • Output 2: Current audio playback CHOP (store_output — latest turn’s audio at 24 kHz)
  • Output 3: Full session audio CHOP (full_audio — accumulated audio across all turns)
  • Output 4: Text output DAT (content from the output_text_content tool, when enabled)

Tool Integration

Section titled “Tool Integration”

Gemini Live consumes tools from connected LOPs operators using the same pattern as the Agent LOP. It does not expose a GetTool() method itself.

Connecting External Tools

Section titled “Connecting External Tools”
  1. On the Tools page, enable ‘Use LOP Tools’.
  2. In the ‘External Op Tools’ sequence, add a block and drag the tool operator into the ‘OP’ field.
  3. Set ‘Mode’ per tool:
    • enabled — the model waits for the tool result before continuing (blocking).
    • enabled_nonblocking — the tool runs without pausing the conversation.
    • disabled — the tool is skipped.
  4. Start the conversation. The model will call tools as needed and incorporate the results into its spoken response.

Built-in Tools

Section titled “Built-in Tools”
  • Allow Model to Stop Conversation: When enabled, the model receives a stop_current_conversation tool it can call when the user says goodbye. The stop is deferred until the current turn completes so the model can finish speaking.
  • Output Text (out4): When enabled, the model receives an output_text_content tool to display text (code snippets, data) in the fourth output DAT without reading it aloud.
  • Enable Google Search (built in): Adds Google Search grounding so the model can look up current information during the conversation.

Usage Examples

Section titled “Usage Examples”

Basic Voice Conversation

Section titled “Basic Voice Conversation”
  1. Enter your Google API key on the Config page (or let ChatTD’s Key Manager supply it).
  2. On the Gemini Live page, choose a ‘Model’ and a ‘Voice’.
  3. Optionally enter a ‘System Prompt’ to set the assistant’s personality.
  4. Pulse ‘Start’. The status indicators will show the connection is live.
  5. Speak into your microphone. The model responds with voice in real time.
  6. Pulse ‘Stop’ to end the conversation.

Sending Video Frames

Section titled “Sending Video Frames”
  1. On the Image page, enable ‘Enable Image Input’.
  2. Reference a TOP in the ‘TOP’ parameter (this feeds frame_null inside the component).
  3. Set ‘Image Send Mode’:
    • Stream (Continuous) sends frames at the interval set in ‘Stream Interval (sec)’.
    • Pulse (Manual) sends a single frame when you pulse ‘Send Image’.
    • Pulse + Send on Start sends one frame automatically when the conversation connects, then manual pulses afterward.
  4. Choose a resolution preset or select ‘Custom’ and set width/height.
  5. Start the conversation. The model can now see and discuss what is in the video feed.

Resuming a Previous Session

Section titled “Resuming a Previous Session”
  1. On the History page, enable ‘Enable Session History’.
  2. Select a saved session from the ‘Session to Load’ menu, then pulse ‘Load Session’. The conversation table populates with the previous messages.
  3. If the session is recent enough (within 12 hours), ‘Resume Session’ is automatically enabled and the Google session resumption handle is restored.
  4. Pulse ‘Start’. The conversation continues with full context from the previous session.

Turn Management

Section titled “Turn Management”

The ‘Turn Management Mode’ on the Gemini Live page controls how user speech is detected:

  • Auto VAD: Audio streams continuously and the server detects speech start/end automatically. Fine-tune detection on the Config page under ‘Configure VAD’ (start/end sensitivity, prefix padding, silence duration).
  • Push to Talk: Audio is only sent while the ‘Push to Talk’ toggle is active. Ideal for noisy environments.
  • Hybrid: Auto VAD runs normally, but the ‘Push to Talk’ toggle can override it for manual control when needed.

Playback

Section titled “Playback”

The Playback page controls audio output hardware:

  • Driver: Default system driver or ASIO for low-latency output.
  • Device: Select the specific audio output device.
  • Volume: Adjust playback volume (0.0 to 1.0).
  • Pulse ‘Clear Audio Buffers’ to flush all generated audio from memory and the output CHOPs.

Troubleshooting

Section titled “Troubleshooting”
  • “google-genai not installed” popup on load: Pulse ‘Install/Update google-genai’ on the Config page, then restart TouchDesigner.
  • Connection drops with 1011 errors: Enable ‘Enable Auto-Reconnect’ on the Config page. Set a reasonable ‘Reconnect Delay’ and ‘Max Reconnect Attempts’. The operator will automatically resume the session after a server error.
  • Session resumption fails (1008 error): The session handle has expired. The operator clears the stale handle automatically and starts a fresh conversation on the next attempt.
  • No audio output: Verify the correct ‘Device’ is selected on the Playback page and ‘Active’ is enabled under Audio Device Settings.

Parameters

Section titled “Parameters”

Gemini Live

Section titled “Gemini Live”
Status (Status) op('gemini_live').par.Status Str
Default:
"" (Empty String)
Live Connection (Statusconnected) op('gemini_live').par.Statusconnected Toggle

Connection status.

Default:
False
Conversation Status (Statusconversationactive) op('gemini_live').par.Statusconversationactive Toggle

Conversation status.

Default:
False
Start (Start) op('gemini_live').par.Start Pulse
Default:
False
Pause (Pause) op('gemini_live').par.Pause Pulse

Pause the current conversation. Will stop and save the session, then prepare for resumption.

Default:
False
Stop (Stop) op('gemini_live').par.Stop Pulse
Default:
False
Resume Session (Resumesession) op('gemini_live').par.Resumesession Toggle

When enabled, Start will resume the loaded session. When disabled, Start will begin a fresh conversation with empty table and no Google session resumption.

Default:
False
Model (Model) op('gemini_live').par.Model StrMenu

Google Live API model for conversation. This is an editable menu - you can type a custom model ID if newer models become available. Available models: • Gemini 2.0 Flash Live (gemini-2.0-flash-live-001): The standard Live API model optimized for real-time bidirectional voice and video interactions. Features include: - Input: Audio, video, and text - Output: Text and audio - 1M token context window - Function calling support - Google Search grounding - Code execution capabilities - Low-latency streaming - Knowledge cutoff: August 2024 • Gemini 2.5 Flash Native Audio (gemini-2.5-flash-native-audio-preview-12-2025): Native audio model for natural conversations: - Input: Audio, video, and text - Output: Audio and text (interleaved) - High quality, natural conversational audio - Thinking support (configurable via thinkingBudget) - 128K token context window - Function calling support - Google Search grounding - Knowledge cutoff: January 2025 For more model details, see: https://ai.google.dev/gemini-api/docs/models#live-api

Default:
"" (Empty String)
Menu Options:
  • Gemini 2.0 Flash Live (gemini-2.0-flash-live-001)
  • Gemini 2.5 Flash Native Audio (gemini-2.5-flash-native-audio-preview-12-2025)
Voice (Voice) op('gemini_live').par.Voice Menu

Voice to use for Google Live API audio responses.

Default:
Puck
Options:
Puck, Charon, Kore, Fenrir, Aoede, Leda, Orus, Zephyr
System Prompt (Systemprompt) op('gemini_live').par.Systemprompt Str

System prompt for the conversation.

Default:
"" (Empty String)
Turn Management Mode (Turnmode) op('gemini_live').par.Turnmode Menu

Turn management mode controls how user speech is detected and sent to the API: • Auto VAD: Continuous audio streaming with server-side Voice Activity Detection. The API automatically detects when you start and stop speaking. • Push to Talk: Manual control via Push to Talk toggle. Audio is only sent when toggle is active. Ideal for noisy environments or precise control. • Hybrid: Combines automatic VAD with Push to Talk override. VAD works normally, but Push to Talk can force activity start/stop for manual control when needed.

Default:
auto_vad
Options:
auto_vad, push_to_talk, hybrid
Push to Talk (Pushtotalk) op('gemini_live').par.Pushtotalk Toggle

Toggle to control when audio is sent (works in Push to Talk, Manual Activity, and Hybrid modes).

Default:
False
Image Send Mode (Imagemode) op('gemini_live').par.Imagemode Menu

Stream: Send frames continuously at specified interval. Pulse (Manual): Send single frame when triggered. Pulse + Send on Start: Manual pulse + automatically send frame when conversation starts.

Default:
pulse
Options:
stream, pulse, pulse_on_connect
Send Image (Sendimage) op('gemini_live').par.Sendimage Pulse

Pulse to send a single video frame (only works in Pulse mode).

Default:
False
TOP (Top) op('gemini_live').par.Top TOP
Default:
"" (Empty String)
Send Text (Sendtext) op('gemini_live').par.Sendtext Pulse

Pulse to send text content from text_input DAT to the conversation.

Default:
False
DAT (Dat) op('gemini_live').par.Dat DAT
Default:
"" (Empty String)
Conversation Table Update (Conversationupdate) op('gemini_live').par.Conversationupdate Menu

Determines when the main conversation table is updated. "Live Transcript" updates as text arrives. "On Turn Complete" updates only after each conversational turn.

Default:
live
Options:
live, on_turn_complete

Tools

Section titled “Tools”
Gemini Live Tools Header
Allow Model to Stop Conversation (Allowmodelstop) op('gemini_live').par.Allowmodelstop Toggle

If enabled, the model will be given a tool to stop the current conversation.

Default:
False
Output Text (out4) (Outputtext) op('gemini_live').par.Outputtext Toggle

Use this tool to display text content (like code, data, or long text) without speaking it aloud. This will appear in out4.

Default:
False
Enable Google Search (built in) (Enablegrounding) op('gemini_live').par.Enablegrounding Toggle

Allow the model to use Google Search to improve accuracy and recency.

Default:
False
Use LOP Tools Header
Use LOP Tools (Usetools) op('gemini_live').par.Usetools Toggle

Enable external tool operators via Tool sequence blocks (similar to Agent operator).

Default:
False
External Op Tools (Tool) op('gemini_live').par.Tool Sequence
Default:
0
Mode (Tool0mode) op('gemini_live').par.Tool0mode Menu
Default:
enabled
Options:
enabled, enabled_nonblocking, disabled
OP (Tool0op) op('gemini_live').par.Tool0op OP
Default:
"" (Empty String)

Image

Section titled “Image”
Enable Image Input (Enableimage) op('gemini_live').par.Enableimage Toggle

Enable video frame input to the Live API. Requires frame_null TOP inside component.

Default:
False
Resolution [ fit outside ] (Imageresolution) op('gemini_live').par.Imageresolution Menu

Resolution handling for video frames: • Use TOP Resolution: Send frames at the resolution of frame_null TOP • 1024x1024: Square format (recommended max for Live API) • 512x512: Square format (good balance of quality/bandwidth) • 256x256: Square format (low bandwidth) • 1920x1080: Full HD 16:9 • 1280x720: HD 16:9 • 854x480: SD 16:9 • 1024x768: 4:3 aspect ratio • 800x600: 4:3 aspect ratio • Custom: Use custom width/height parameters

Default:
use_top
Options:
use_top, 1024x1024, 512x512, 256x256, 1920x1080, 1280x720, 854x480, 1024x768, 800x600, custom
Stream Interval (sec) (Streaminterval) op('gemini_live').par.Streaminterval Float

Time between frames when in Stream mode (seconds). Lower = more frames, higher bandwidth.

Default:
0.0
Range:
0.1 to 10
Slider Range:
0.1 to 10
Custom Width (Customwidth) op('gemini_live').par.Customwidth Int

Custom frame width (only used when Video Resolution is set to Custom).

Default:
0
Range:
64 to 2048
Slider Range:
64 to 2048
Custom Height (Customheight) op('gemini_live').par.Customheight Int

Custom frame height (only used when Video Resolution is set to Custom).

Default:
0
Range:
64 to 2048
Slider Range:
64 to 2048
Media Resolution (Mediaresolution) op('gemini_live').par.Mediaresolution Menu

Official API media resolution setting. Controls quality vs bandwidth trade-off for video content.

Default:
low
Options:
low, default, high

Playback

Section titled “Playback”
Audio Device Settings Header
Active (Audioactive) op('gemini_live').par.Audioactive Toggle
Default:
True
Driver (Driver) op('gemini_live').par.Driver Menu
Default:
default
Options:
default, asio
Device (Device) op('gemini_live').par.Device Menu
Default:
default
Options:
default, {0.0.0.00000000}.{d7b929aa-ec27-4f96-bd39-78d6a8c2044a}||Out_1-2_(MOTU_M_Series)||1, {0.0.0.00000000}.{044f8ef8-f1ad-4655-90d6-0aef7b713b78}||Voicemeeter_AUX_Input_(VB-Audio_Voicemeeter_VAIO)||2, {0.0.0.00000000}.{170cc7c6-264f-46f4-a652-4c80058e49d2}||LS27A70_(NVIDIA_High_Definition_Audio)||3, {0.0.0.00000000}.{2162b344-60a4-4dda-a068-a0887826a518}||Voicemeeter_VAIO3_Input_(VB-Audio_Voicemeeter_VAIO)||4, {0.0.0.00000000}.{25b313f1-54e3-4bae-b661-e07474413cab}||Voicemeeter_In_5_(VB-Audio_Voicemeeter_VAIO)||5, {0.0.0.00000000}.{34b7624e-63b1-49fc-93dc-1d03ca1dd600}||CABLE-B_In_16ch_(VB-Audio_Virtual_Cable_B)||6, {0.0.0.00000000}.{372fb62b-07aa-4580-b062-2c6adba187e7}||Out_3-4_(MOTU_M_Series)||7, {0.0.0.00000000}.{394e3e0c-eac4-4fc9-ad76-1600f0cb570b}||CABLE-A_Input_(VB-Audio_Virtual_Cable_A)||8, {0.0.0.00000000}.{6b4330e7-1895-4e98-8c9b-f21172031db3}||F13NA_(NVIDIA_High_Definition_Audio)||9, {0.0.0.00000000}.{95545e0a-9e74-42cc-9835-fa5b33914d6c}||Voicemeeter_In_2_(VB-Audio_Voicemeeter_VAIO)||10, {0.0.0.00000000}.{9f9a792f-c06d-4923-9f4b-da5200378f26}||LEN_P32u-10_(NVIDIA_High_Definition_Audio)||11, {0.0.0.00000000}.{a4936fe7-9e56-4176-9789-f343437419f2}||Voicemeeter_In_4_(VB-Audio_Voicemeeter_VAIO)||12, {0.0.0.00000000}.{a720258f-2455-411a-a701-1f8ecee9d3d6}||Voicemeeter_In_3_(VB-Audio_Voicemeeter_VAIO)||13, {0.0.0.00000000}.{ad8837d6-c905-4a72-adcb-7018ee7baab3}||CABLE_Input_(VB-Audio_Virtual_Cable)||14, {0.0.0.00000000}.{b6526f47-8c31-48d5-8b30-489196c56a6b}||Headphones_(iLoud_Micro-Monitor)||15, {0.0.0.00000000}.{b8fcdf16-5164-47d2-80aa-0f80ce4bd0b5}||Voicemeeter_Input_(VB-Audio_Voicemeeter_VAIO)||16, {0.0.0.00000000}.{c6e56f21-72eb-46b0-9761-7ba5c2286004}||Voicemeeter_In_1_(VB-Audio_Voicemeeter_VAIO)||17, {0.0.0.00000000}.{d8576751-f212-4fa7-8cb5-0f825e64c87e}||CABLE-B_Input_(VB-Audio_Virtual_Cable_B)||18, {0.0.0.00000000}.{fddd6891-fc7b-4afa-8c26-9246692a19f0}||CABLE-A_In_16ch_(VB-Audio_Virtual_Cable_A)||19
Volume (Volume) op('gemini_live').par.Volume Float
Default:
1.0
Range:
0 to 1
Slider Range:
0 to 1
Clear Audio Buffers (Clearaudio) op('gemini_live').par.Clearaudio Pulse

Clears all generated audio from memory and from the output CHOPs (store_output and full_audio).

Default:
False

History

Section titled “History”
Enable Session History (Enablesessionhistory) op('gemini_live').par.Enablesessionhistory Toggle

Save and load session history to/from JSON files for conversation recall.

Default:
False
Save Session (Savesession) op('gemini_live').par.Savesession Pulse

Manually save current session to history file.

Default:
False
Load Session (Loadsession) op('gemini_live').par.Loadsession Pulse

Load a previous session from history. Shows file dialog.

Default:
False
List Sessions (Listsessions) op('gemini_live').par.Listsessions Pulse

Display all saved sessions with detailed metadata.

Default:
False
Session to Load (Sessiontoload) op('gemini_live').par.Sessiontoload Menu

Specify session ID for targeted loading.

Default:
"" (Empty String)
Options:
List All Sessions (Listallsessions) op('gemini_live').par.Listallsessions Toggle
Default:
False

Config

Section titled “Config”
API Key (Apikey) op('gemini_live').par.Apikey Str

Google API key for authentication.

Default:
"" (Empty String)
Install/Update google-genai (Installgooglegenai) op('gemini_live').par.Installgooglegenai Pulse

Installs or updates the 'google-genai' Python package using the project's Python Manager. This is required for the operator to function.

Default:
False
Enable User Transcription (Enableusertranscription) op('gemini_live').par.Enableusertranscription Toggle

Enable transcription of user speech for conversation memory.

Default:
False
Enable Session Resumption (Enablesessionresumption) op('gemini_live').par.Enablesessionresumption Toggle

Use session resumption for continuous conversation.

Default:
False
Enable Context Compression (Enablecontextcompression) op('gemini_live').par.Enablecontextcompression Toggle

Enable context window compression for longer sessions.

Default:
False
Audio Send Interval (sec) (Audiosendinterval) op('gemini_live').par.Audiosendinterval Float

How often to send audio chunks to Google.

Default:
0.0
Range:
0.05 to 0.5
Slider Range:
0.05 to 0.5
Configure VAD (Enablevadconfig) op('gemini_live').par.Enablevadconfig Toggle

Enable custom Voice Activity Detection (VAD) settings. If False, API defaults are used.

Default:
False
VAD Start Sensitivity (Startofspeechsensitivity) op('gemini_live').par.Startofspeechsensitivity Menu

Sensitivity for detecting the start of speech. Requires "Configure VAD" to be enabled.

Default:
low
Options:
low, medium, high, highest
VAD End Sensitivity (Endofspeechsensitivity) op('gemini_live').par.Endofspeechsensitivity Menu

Sensitivity for detecting the end of speech. Requires "Configure VAD" to be enabled.

Default:
low
Options:
low, medium, high, highest
VAD Prefix Padding (ms) (Prefixpaddingms) op('gemini_live').par.Prefixpaddingms Int

Milliseconds of audio to include before detected start of speech. Requires "Configure VAD".

Default:
0
Range:
0 to 500
Slider Range:
0 to 500
VAD Silence Duration (ms) (Silencedurationms) op('gemini_live').par.Silencedurationms Int

Milliseconds of silence to detect as end of speech. Requires "Configure VAD".

Default:
0
Range:
100 to 5000
Slider Range:
100 to 5000
Language Code (BCP-47) (Languagecode) op('gemini_live').par.Languagecode Str

Optional. Sets response language (e.g., en-US, es-ES). Overrides voice/model default if set. May not apply to all models.

Default:
"" (Empty String)
Enable Auto-Reconnect (Enableautoreconnect) op('gemini_live').par.Enableautoreconnect Toggle

Automatically reconnect and resume session when connection is lost due to server errors (like 1011 internal errors).

Default:
False
Reconnect Delay (sec) (Reconnectdelay) op('gemini_live').par.Reconnectdelay Float

How long to wait before attempting to reconnect after a connection failure.

Default:
0.0
Range:
1 to 30
Slider Range:
1 to 30
Max Reconnect Attempts (Maxreconnectattempts) op('gemini_live').par.Maxreconnectattempts Int

Maximum number of reconnection attempts before giving up. Set to 0 for unlimited attempts.

Default:
0
Range:
1 to 10
Slider Range:
1 to 10
Current Reconnect Attempts (Reconnectattempts) op('gemini_live').par.Reconnectattempts Int

Current number of reconnection attempts (read-only status).

Default:
0
Range:
0 to 1
Slider Range:
0 to 1

Changelog

Section titled “Changelog”
v2.3.32026-03-01
  • Update Live API models to current IDs (remove deprecated gemini-2.5-flash-preview-native-audio-dialog and gemini-2.5-flash-exp-native-audio-thinking-dialog) - Add gemini-2.5-flash-native-audio-preview-12-2025 as current native audio model - Change Model parameter from menu to strmenu for custom model ID entry - Default model set to gemini-2.5-flash-native-audio-preview-12-2025
  • Minor change - added reset - does not qualify for release
  • Initial commit
v2.3.2

🔧 Major Fixes

Fixed MCP Server Integration Issues

  • Tool Schema Validation: Resolved LiveConnectConfig validation errors that prevented MCP server tools from working
  • Parameter Format Conversion: Fixed tool parameter type conversion issues causing external tool failures
  • Schema Compatibility: Enhanced schema cleaning to handle Gemini Live API requirements

🛠 Technical Improvements

Schema Processing Enhancements

  • Added automatic removal of additionalProperties and $schema fields from tool schemas
  • Implemented missing items field detection and auto-repair for array-type parameters
  • Fixed Any type compatibility issues by defaulting to str type for Live API

Parameter Transformation System

  • Enhanced Array Handling: Added support for multiline string to array conversion (newline-separated)
  • Improved Type Inference: Parameters with items field now correctly inferred as array type even without explicit type: 'array'
  • Better Format Support: Enhanced comma-separated string parsing and single-value array wrapping
  • Added comprehensive parameter validation with helpful warning messages

Tool Argument Processing

  • Fixed string-to-array conversion for tool parameters like insert_column values
  • Added support for multiline string parsing (common in tool_dat operations)
  • Improved type coercion for integer, number, boolean, and object parameters
  • Enhanced error reporting for parameter format mismatches

🐛 Bug Fixes

  • Tool_DAT Integration: Fixed insert_column and other tool_dat operations failing due to parameter format issues
  • MCP Server Tools: Resolved "missing field" errors preventing MCP tools from loading
  • Live API Compatibility: Fixed callable function format issues in LiveConnectConfig
  • Schema Validation: Eliminated pydantic validation errors for external tool schemas

📝 Developer Notes

All changes were made exclusively to GoogleVoice2VoiceEXT.py:

  • Enhanced _clean_parameters_recursive() method for better schema cleaning
  • Improved transform_tool_arguments() method with array type inference
  • Added _validate_transformed_arguments() method for parameter validation
  • Updated type mappings to use str instead of Any for Live API compatibility

✅ Verification

  • MCP server tools now integrate successfully with Gemini Live API
  • Tool_DAT operations (insert_column, replace_all_table, etc.) working correctly
  • Parameter format conversions handle various input formats automatically
  • Enhanced logging for debugging tool parameter issues

Impact: This release resolves all major MCP integration issues and tool parameter format problems, enabling full compatibility between external tools and the Gemini Live API.

v2.3.12025-09-01

improved handling of nested arrays to flatten them into strings for gemini live and avoid the 2 depth limit for items in gemini live api funcitoncalling schema format.

v2.3.02025-08-18

New Features

  • Added session history table with readable labels (date, duration, message count, resumable status)
  • Added pause conversation functionality
  • Added resume_status boolean column for easy session filtering
  • Shortened parameter names: Start, Stop, Pause (instead of Startconversation, etc.)

Session Management Improvements

  • Smart session resumption (only attempts recent sessions < 12 hours old)
  • Always creates new session IDs even when resuming (preserves conversation history)
  • Sessions sorted newest first in history table
  • Compact label format: "Aug18 18:43 25msgs 18s" with "View Only" indicator for old sessions

Bug Fixes

  • Fixed final message capture when assistant ends conversation (added 500ms + 100ms delays)
  • Fixed session age parsing for both timestamp and ISO string formats
  • Fixed table clearing logic to preserve loaded conversation history during resumption
  • Improved error handling for expired session handles (1008 errors)

Technical Changes

  • Enhanced session loading with age validation and resumption handle checks
  • Improved logging transparency for session operations
  • Model stop tool properly triggers session saving
v2.2.12025-08-01
  • Fixed MCP tool integration: Array parameters now include required 'items' field in schema conversion
  • Resolved "missing fie" error when using MCP tools with array parameters like search_terms
v2.2.02025-07-13

Added / fixed the Tool mode so that theree is the ability to set per op tool block or non blocking mode rather than needing to specify that in the GetTool definiotn since it is gemini specific. also allows for more specifity to choose to wait or not wait for the tool to complete - whether or not the agent waits still might depend on the system prompt or instructions or nature of the tool used.

v2.1.02025-06-30
  • fixed the conversation table updating correctly and added live mode so the table updates live. (par.Conversationupdate [menu: 'live', 'on_turn_complete'])
  • added Playback parameter page that controls the palyback of the agent audio + added a clear audio button to clear the audio buffers.
v2.0.02025-06-25

added video + text input

parity of tools between agent op

added textoutput internal tool.

v1.0.02025-05-25

Initial release

Section titled “Related Operators”