> ## Documentation Index
> Fetch the complete documentation index at: https://docs.zeroruntime.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Changelog

> What changed in each release of the Zero Runtime Python and Node.js SDKs.

Release notes for the Zero Runtime SDKs, newest first. Use the view selector to switch between the Python and Node.js SDKs.

**Python**

Upgrade with `pip install -U zeroruntime`. See the [Python SDK](https://docs.zeroruntime.ai/runtime-sdk/sdks/python.md) page for setup.

**0.2.1**

Subagents, agent workflows, Mistral AI and TLS settings

Hand calls between agents with `transfer_to()`, sequence agents from YAML, and run the SDK behind a TLS-inspecting proxy.

### Agents

| Feature | What it does | Docs |
| - | - | - |
| `transfer_to()` | Hands a call to another agent. Works with agents that have workflows and with plain handoffs. | [Subagents](https://docs.zeroruntime.ai/runtime-sdk/build/subagents.md) |
| Agent workflows | Define agent sequencing in a YAML file. | [Workflows and steps](https://docs.zeroruntime.ai/runtime-sdk/build/workflows-and-steps.md) |
| `agent.userdata` | Read call metadata from inside an agent. | [Workflows and steps](https://docs.zeroruntime.ai/runtime-sdk/build/workflows-and-steps.md) |
| `agent.agent_workflow` | Read the subagents configured for the current agent. | [Subagents](https://docs.zeroruntime.ai/runtime-sdk/build/subagents.md) |

### Plugins

| Class | Purpose | Docs |
| - | - | - |
| `MistralAISTT` | Speech-to-text | [Mistral AI STT](https://docs.zeroruntime.ai/runtime-sdk/plugins/stt/mistralai.md) |
| `MistralAILLM` | LLM | [Mistral AI LLM](https://docs.zeroruntime.ai/runtime-sdk/plugins/llm/mistralai.md) |
| `MistralAITurnDetectionConfig` | Turn detection settings | [Mistral AI STT](https://docs.zeroruntime.ai/runtime-sdk/plugins/stt/mistralai.md) |

### TLS and networking

Both variables are read from the environment before the SDK connects.

```bash theme={null}
# Trust a custom CA, for example behind a TLS-inspecting proxy
export ZERORUNTIME_CA_FILE=/path/to/ca.pem

# Skip certificate checks for HTTP and WebSocket calls (local debugging only)
export ZERORUNTIME_TLS_VERIFY=false
```

> [!WARNING]
>
> `ZERORUNTIME_TLS_VERIFY=false` does not affect gRPC. gRPC connections always verify certificates.

**0.2.0**

New speech plugins, prosody control and faster session start

Adds four speech providers and two speech-to-speech integrations, and starts sessions in a single request.

### Plugins

| Class | Purpose | Docs |
| - | - | - |
| `MistralAITTS` | Expressive multilingual speech with Mistral's Voxtral voices | [Mistral AI TTS](https://docs.zeroruntime.ai/runtime-sdk/plugins/tts/mistralai.md) |
| `FishAudioSTT`, `FishAudioTTS` | Speech-to-text and text-to-speech. Direct API key or inference. | [STT](https://docs.zeroruntime.ai/runtime-sdk/plugins/stt/fishaudio.md), [TTS](https://docs.zeroruntime.ai/runtime-sdk/plugins/tts/fishaudio.md) |
| `GradiumSTT`, `GradiumTTS` | Speech-to-text and text-to-speech. Direct API key. | [STT](https://docs.zeroruntime.ai/runtime-sdk/plugins/stt/gradium.md), [TTS](https://docs.zeroruntime.ai/runtime-sdk/plugins/tts/gradium.md) |
| `AssemblyAITTS` | Text-to-speech. Direct API key or inference. | [AssemblyAI TTS](https://docs.zeroruntime.ai/runtime-sdk/plugins/tts/assemblyai.md) |
| `PhonicRealtime`, `PhonicRealtimeConfig` | Speech-to-speech | [Phonic](https://docs.zeroruntime.ai/runtime-sdk/plugins/realtime/phonic.md) |
| `RimeRealtime`, `RimeRealtimeConfig` | Speech-to-speech using Rime's Prism model | [Rime](https://docs.zeroruntime.ai/runtime-sdk/plugins/realtime/rime.md) |

**0.1.9**

Interrupt phrases, knowledge base retrieval, OpenAI and Anthropic

Adds OpenAI and Anthropic support, direct knowledge base search, and short-phrase interruptions.

### Turn-taking and telephony

* **Interrupt phrases.** Add `interrupt_phrases` to `InterruptSpec` and `InterruptConfig` so short replies like "stop" interrupt the agent even below the minimum word threshold. See [VAD and interruptions](https://docs.zeroruntime.ai/runtime-sdk/build/turn-detection-and-interruptions/vad-and-interruptions.md) and the [pipeline API reference](https://docs.zeroruntime.ai/api-reference/python/core/pipeline.md).
* **Warm transfer.** `WarmTransferConfig` now sets consultation rooms, hold messages and failure messages. See [Warm transfer](https://docs.zeroruntime.ai/runtime-sdk/build/telephony/warm-transfer.md).

### Knowledge base

* **`retrieve`.** A new async method searches knowledge base entries directly. See [RAG](https://docs.zeroruntime.ai/runtime-sdk/build/tools-and-capabilities/rag.md).
* **Retrieval metrics hook.** Monitor knowledge base searches through [observability hooks](https://docs.zeroruntime.ai/runtime-sdk/build/configure-a-pipeline/observability-hooks.md).

### Providers

| Provider | Classes | Notes | Docs |
| - | - | - | - |
| OpenAI | `OpenAILLM`, `OpenAISTT`, `OpenAITTS`, `OpenAIRealtime`, `OpenAILive` | Also available through managed inference | [LLM](https://docs.zeroruntime.ai/runtime-sdk/plugins/llm/openai.md), [STT](https://docs.zeroruntime.ai/runtime-sdk/plugins/stt/openai.md), [TTS](https://docs.zeroruntime.ai/runtime-sdk/plugins/tts/openai.md), [Realtime](https://docs.zeroruntime.ai/runtime-sdk/plugins/realtime/openai.md) |
| Anthropic | `AnthropicLLM` | Defaults to `claude-haiku-5-5` with prompt caching on. Also available through managed inference. | [Anthropic LLM](https://docs.zeroruntime.ai/runtime-sdk/plugins/llm/anthropic.md) |
| GPT-Live | `OpenAILive` | Ultrafast service tier. Default model is now `gpt-5.6-luna`. | [GPT-Live](https://docs.zeroruntime.ai/runtime-sdk/plugins/duplex/gpt-live.md) |

See [Inference Gateway](https://docs.zeroruntime.ai/runtime-sdk/plugins/inference/zero-runtime.md) for managed inference.

**0.1.8**

IDE experience and Deepgram TTS updates

* **IDE support.** Better autocomplete and type hints in the Python client.
* **Deepgram TTS.** Updated parameters for expressivity and connection settings. See [Deepgram TTS](https://docs.zeroruntime.ai/runtime-sdk/plugins/tts/deepgram.md).
* **Fix.** `Room` now handles the default target redaction correctly.

> [!WARNING]
>
> `KnowledgeBase.report_unanswered` is deprecated.

**0.1.7**

Tool cancellation controls and PII redaction

Gives you control over what happens to running tools when a caller interrupts.

### Tools

* **`cancel_on_interrupt`.** Set it on `@function_tool` to choose whether a caller interruption cancels a running tool. Tools can keep running after an interruption. See [Function tools](https://docs.zeroruntime.ai/runtime-sdk/build/tools-and-capabilities/function-tools.md).
* **Speech from tools.** Call `say()` and `reply()` inside a tool to speak while it runs.
* **Session cleanup.** Running tools are cancelled when the session ends.

### Privacy

* **PII redaction.** Redact sensitive data from transcripts and recordings. See [PII redaction](https://docs.zeroruntime.ai/runtime-sdk/build/observability-and-monitoring/pii-redaction.md).

### Changed defaults

| Area | Change | Docs |
| - | - | - |
| Agent ID | Read from an environment variable. The `agent_id` parameter is deprecated. | n/a |
| Google LLM | Default model is now `gemini-3.5-flash-lite` | [Google LLM](https://docs.zeroruntime.ai/runtime-sdk/plugins/llm/google.md) |
| Knowledge base and duplex | Updated support for both | n/a |

**0.1.6**

OpenAI Duplex, context appending and knowledge base

### New

* **Knowledge base.** `KnowledgeBase` support lets agents retrieve information from configured knowledge bases. See [RAG](https://docs.zeroruntime.ai/runtime-sdk/build/tools-and-capabilities/rag.md).
* **OpenAI Duplex.** Support for `gpt-live-1`. See [GPT-Live](https://docs.zeroruntime.ai/runtime-sdk/plugins/duplex/gpt-live.md).
* **`context_append`.** New field on `ClientMessage`.
* **Duplex context.** Append instructions, thinking and commentary to duplex models during a session. See [Context control](https://docs.zeroruntime.ai/runtime-sdk/build/context-management/context-control.md).
* **OpenAI configuration.** New options for the OpenAI backend and the delegate LLM.

### Changed

> [!WARNING]
>
> Credential resolution was refactored. The alternate credentials function is removed. Use the new method for setting provider credentials instead.

**0.1.5**

Playground events and Echo Omni

* **`AgentEventsOptions`.** Pass it to `Room` to send interim transcripts and realtime metrics to the playground.
* **Echo Omni.** Enable it with the turn detector below. See [Turn detection](https://docs.zeroruntime.ai/runtime-sdk/build/turn-detection-and-interruptions/turn-detection.md).

```python theme={null}
TurnDetector(model="echo-omni", language="en")
```

**0.1.4**

Developer experience improvements

Smoother developer experience for:

* [Fallback adapter](https://docs.zeroruntime.ai/runtime-sdk/build/configure-a-pipeline/fallback-adapter.md)
* [DTMF](https://docs.zeroruntime.ai/runtime-sdk/build/telephony/dtmf.md)
* [Voicemail detection](https://docs.zeroruntime.ai/runtime-sdk/build/telephony/voicemail-detection.md)
* Pub/sub

**0.1.3**

Realtime pipeline arguments and stability fixes

* **Realtime models.** Supported in the LLM pipeline arguments. See [Pipeline modes](https://docs.zeroruntime.ai/runtime-sdk/build/configure-a-pipeline/modes.md).

### Fixes

* Fixed agent cleanup issues that caused unbounded memory growth.
* Fixed tool timeout handling during warm transfers.
* Fixed stream reading after warm-transfer phase events and call-transfer results.

**0.1.2**

Inference service and docs tooling

* **Inference.** Enhancements to the Zero Runtime Inference Service. See [Inference Gateway](https://docs.zeroruntime.ai/runtime-sdk/plugins/inference/zero-runtime.md).
* **Fix.** The `hangup` method now works as expected.
* **Docs.** Improved SDK docstrings and added API reference generation.

**0.1.1**

Package rename from ZRT to ZeroRuntime

> [!WARNING]
>
> Code written for 0.1.0 does not run unchanged on 0.1.1. Update imports, the CLI command and environment variables as listed below.

| | Before | After |
| - | - | - |
| Python import | `zrt` | `zeroruntime` |
| CLI command | `zrt` | `zeroruntime` |
| Environment variables | `ZRT_*` | `ZERORUNTIME_*` |

The PyPI package name is unchanged. See the [CLI reference](https://docs.zeroruntime.ai/runtime-sdk/deployments/cli-reference.md).

**0.1.0**

Initial Python SDK release

First release of the Zero Runtime Python SDK on PyPI. Start with the [Python SDK](https://docs.zeroruntime.ai/runtime-sdk/sdks/python.md) page.

| Area | What's included | Docs |
| - | - | - |
| Pipelines | Cascading STT → LLM → TTS, and realtime speech-to-speech models | [Pipeline modes](https://docs.zeroruntime.ai/runtime-sdk/build/configure-a-pipeline/modes.md) |
| Context and memory | Context window control, session context, pub/sub messaging | [Context window](https://docs.zeroruntime.ai/runtime-sdk/build/context-management/context-window.md) |
| Multi-agent | Agent handoffs, MCP over stdio and HTTP | [Subagents](https://docs.zeroruntime.ai/runtime-sdk/build/subagents.md), [MCP](https://docs.zeroruntime.ai/runtime-sdk/build/tools-and-capabilities/mcp.md) |
| Fallbacks | Automatic failover across STT, LLM and TTS providers | [Fallback adapter](https://docs.zeroruntime.ai/runtime-sdk/build/configure-a-pipeline/fallback-adapter.md) |
| Tools | `@function_tool` turns Python functions into agent tools, with speech fillers while they run | [Function tools](https://docs.zeroruntime.ai/runtime-sdk/build/tools-and-capabilities/function-tools.md) |
| Telephony | DTMF, call transfers, SIP room options | [Telephony](https://docs.zeroruntime.ai/runtime-sdk/build/telephony/overview.md) |
| Turn-taking | Turn detection, VAD, denoising, configurable interruptions | [Turn detection and interruptions](https://docs.zeroruntime.ai/runtime-sdk/build/turn-detection-and-interruptions/overview.md) |
| Audio and media | Background audio, thinking audio, pronunciation rules, camera input, avatars | [Modalities](https://docs.zeroruntime.ai/runtime-sdk/build/modalities/overview.md) |
| Observability | Session recording, OTLP traces, metrics, logs | [Observability](https://docs.zeroruntime.ai/runtime-sdk/build/observability-and-monitoring/melt/overview.md) |
| Deployment | The `zrt` CLI for installation and deployments | [Deployments](https://docs.zeroruntime.ai/runtime-sdk/deployments/overview.md) |

**Node.js**

Install or upgrade with `npm install @zeroruntime/js-sdk@latest`.

**0.2.0**

Workflows, subagents, new providers and TLS settings

Adds workflows and `transfer_to` for multi-agent calls, nine new provider classes, and direct knowledge base search.

### Agents and workflows

| Feature | What it does | Docs |
| - | - | - |
| `transfer_to` | Hands a call to any agent. Agents with workflow steps run on the runtime's workflow engine. | [Subagents](https://docs.zeroruntime.ai/runtime-sdk/build/subagents.md) |
| `Step`, `Workflow`, `Workflow.from_yaml` | Define agent sequencing in code or from a YAML file. | [Workflows and steps](https://docs.zeroruntime.ai/runtime-sdk/build/workflows-and-steps.md) |
| `instructions` | Now optional. The default agent name is the subclass name. | [Agent and session](https://docs.zeroruntime.ai/api-reference/node-js/core/agent-and-session.md) |
| `on_enter` | Strings returned from `on_enter` are spoken. | [Agent and session](https://docs.zeroruntime.ai/api-reference/node-js/core/agent-and-session.md) |
| `AgentEvent` | New `agent_handoff` and workflow step events. | [Events](https://docs.zeroruntime.ai/api-reference/node-js/core/events.md) |

### Knowledge base and interruptions

* **`KnowledgeBase.retrieve`.** Search knowledge base entries from a tool. Pair it with the new `knowledge_base` hook. See [RAG](https://docs.zeroruntime.ai/runtime-sdk/build/tools-and-capabilities/rag.md).
* **`InterruptConfig.interrupt_phrases`.** Short replies like "stop" can interrupt the agent. See [VAD and interruptions](https://docs.zeroruntime.ai/runtime-sdk/build/turn-detection-and-interruptions/vad-and-interruptions.md).
* **Warm transfer.** `WarmTransferConfig.room`, `hold_message` and `failure_message` configure the consultation room and transfer messages. See [Warm transfer](https://docs.zeroruntime.ai/runtime-sdk/build/telephony/warm-transfer.md).
* **`Room({ chat_topic })`.** Routes pub/sub messages to the agent and publishes its replies back to the topic. See [Chat](https://docs.zeroruntime.ai/runtime-sdk/build/modalities/text/chat.md).

### Providers

| Class | Purpose | Docs |
| - | - | - |
| `FishAudioSTT`, `FishAudioTTS` | Speech-to-text and text-to-speech | [STT](https://docs.zeroruntime.ai/runtime-sdk/plugins/stt/fishaudio.md), [TTS](https://docs.zeroruntime.ai/runtime-sdk/plugins/tts/fishaudio.md) |
| `GradiumSTT`, `GradiumTTS` | Speech-to-text and text-to-speech | [STT](https://docs.zeroruntime.ai/runtime-sdk/plugins/stt/gradium.md), [TTS](https://docs.zeroruntime.ai/runtime-sdk/plugins/tts/gradium.md) |
| `AssemblyAITTS` | Text-to-speech | [AssemblyAI TTS](https://docs.zeroruntime.ai/runtime-sdk/plugins/tts/assemblyai.md) |
| `PhonicRealtime`, `PhonicRealtimeConfig` | Speech-to-speech | [Phonic](https://docs.zeroruntime.ai/runtime-sdk/plugins/realtime/phonic.md) |
| `RimeRealtime`, `RimeRealtimeConfig` | Speech-to-speech | [Rime](https://docs.zeroruntime.ai/runtime-sdk/plugins/realtime/rime.md) |
| `MistralAISTT` | Speech-to-text with turn detection | [Mistral AI STT](https://docs.zeroruntime.ai/runtime-sdk/plugins/stt/mistralai.md) |
| `MistralAILLM` | LLM | [Mistral AI LLM](https://docs.zeroruntime.ai/runtime-sdk/plugins/llm/mistralai.md) |
| `ZRTSTT` | Speech-to-text through Zero Runtime | [Inference Gateway](https://docs.zeroruntime.ai/runtime-sdk/plugins/inference/zero-runtime.md) |
| `ProsodyConfig` | Speech speed and volume | n/a |

`GeminiLiveConfig` now accepts `null` to disable a feature.

### Sessions

* **Single-request start.** A session starts and connects through one EventStream.
* **Backward compatible.** Older runtimes fall back to `CreateSession`.
* **CLI.** Updated to follow the latest release.

### TLS and networking

```bash theme={null}
# Trust a custom CA, for example behind a TLS-inspecting proxy
export ZERORUNTIME_CA_FILE=/path/to/ca.pem

# Skip certificate checks (local debugging only)
export ZERORUNTIME_TLS_VERIFY=false
```

**0.1.2**

PII redaction and tool cancellation

* **PII redaction.** Redact personally identifiable information in observability data. Redaction adds a `target_was_defaulted` flag and a `without` method for managing targets dynamically. See [PII redaction](https://docs.zeroruntime.ai/runtime-sdk/build/observability-and-monitoring/pii-redaction.md).
* **Tool cancellation.** Running tools are cancelled when an interruption occurs.
* **Agent ID.** Now read from environment variables, which simplifies deployments.
* **CLI.** Updated to version `0.0.8`.
* **Google LLM.** `GoogleLLM` is pinned to a specific version for consistency.

**0.1.1**

Knowledge base in the pipeline and duplex session fixes

* **Knowledge base.** `KnowledgeBase` is now integrated directly into the pipeline. See [RAG](https://docs.zeroruntime.ai/runtime-sdk/build/tools-and-capabilities/rag.md).
* **Nested providers.** New configuration options support nested providers and `OpenAILive` delegation.
* **Duplex models.** Better session lifecycle management.
* **`ZRTLLM`.** Remains the central gateway for LLM calls.
* **STT interim results.** Default interim-result behavior is now consistent across supported STT providers.

**0.1.0**

Initial JavaScript SDK release

First release of `@zeroruntime/js-sdk` on npm. Start with the [Node.js SDK](https://docs.zeroruntime.ai/runtime-sdk/sdks/nodejs.md) page.

> [!NOTE]
>
> Requires Node.js 20.11 or newer. The package is ESM-only and ships complete type declarations. Run TypeScript through `tsx`. JavaScript needs no build step.

| Area | What's included | Docs |
| - | - | - |
| Pipelines | Cascading STT → LLM → TTS, and realtime speech-to-speech. Swap components mid-session. | [Pipeline modes](https://docs.zeroruntime.ai/runtime-sdk/build/configure-a-pipeline/modes.md) |
| Context and memory | Context window control, session context, pub/sub between agents and rooms | [Context window](https://docs.zeroruntime.ai/runtime-sdk/build/context-management/context-window.md) |
| Multi-agent | Mid-session agent handoffs, MCP over stdio and HTTP | [Subagents](https://docs.zeroruntime.ai/runtime-sdk/build/subagents.md), [MCP](https://docs.zeroruntime.ai/runtime-sdk/build/tools-and-capabilities/mcp.md) |
| Fallbacks | Automatic failover across STT, LLM and TTS providers | [Fallback adapter](https://docs.zeroruntime.ai/runtime-sdk/build/configure-a-pipeline/fallback-adapter.md) |
| Tools | `function_tool` registers functions as agent tools | [Function tools](https://docs.zeroruntime.ai/runtime-sdk/build/tools-and-capabilities/function-tools.md) |
| Telephony | DTMF handlers, call transfers, SIP room options | [Telephony](https://docs.zeroruntime.ai/runtime-sdk/build/telephony/overview.md) |
| Turn-taking | Turn detection, VAD, denoising, configurable end-of-utterance and interruptions | [Turn detection and interruptions](https://docs.zeroruntime.ai/runtime-sdk/build/turn-detection-and-interruptions/overview.md) |
| Audio and media | Background audio, thinking audio, pronunciation rules, camera input, avatars | [Modalities](https://docs.zeroruntime.ai/runtime-sdk/build/modalities/overview.md) |
| Observability | Session recording, per-session OTLP traces, metrics, logs | [Observability](https://docs.zeroruntime.ai/runtime-sdk/build/observability-and-monitoring/melt/overview.md) |
| Providers | STT, LLM, TTS, realtime, avatar, VAD and denoising providers | [Plugins](https://docs.zeroruntime.ai/runtime-sdk/plugins/overview.md) |
| Inference gateway | Use supported providers with a Zero Runtime token instead of individual vendor API keys | [Inference Gateway](https://docs.zeroruntime.ai/runtime-sdk/plugins/inference/zero-runtime.md) |
| Deployment | The `zeroruntime` CLI for installation and deployments | [Deployments](https://docs.zeroruntime.ai/runtime-sdk/deployments/overview.md) |


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.