diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index 86b213c..5c8d751 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -2,9 +2,9 @@ name: CI
 
 on:
   push:
-    branches: [main, develop]
+    branches: [master, develop]
   pull_request:
-    branches: [main]
+    branches: [master]
 
 jobs:
   lint-and-test:
@@ -28,12 +28,6 @@ jobs:
           path: ~/.cache/pip
           key: ${{ runner.os }}-pip-${{ matrix.python-version }}-${{ hashFiles('pyproject.toml') }}
 
-      - name: Install build tools
-        run: pip install hatchling grpcio-tools
-
-      - name: Generate gRPC stubs
-        run: python hatch_build.py
-
       - name: Install package with dev dependencies
         run: pip install -e ".[dev]"
 
@@ -62,7 +56,7 @@ jobs:
           python-version: "3.11"
 
       - name: Install build tools
-        run: pip install build hatchling grpcio-tools
+        run: pip install build hatchling
 
       - name: Build wheel and sdist
         run: python -m build
diff --git a/.gitignore b/.gitignore
index fcedf93..97262a1 100644
--- a/.gitignore
+++ b/.gitignore
@@ -16,9 +16,6 @@ venv/
 env/
 .env
 
-# Generated gRPC stubs (regenerate with: make proto)
-livekit/plugins/techmo/_proto/techmo/
-
 # IDEs
 .idea/
 .vscode/
diff --git a/CHANGELOG.md b/CHANGELOG.md
index f3feca0..a9a0a6d 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -4,6 +4,26 @@ All notable changes to this project will be documented here.
 
 Format follows [Keep a Changelog](https://keepachangelog.com/en/1.0.0/).
 
+## [Unreleased]
+
+### Fixed
+- Fresh installs failing on import with `VersionError` (protobuf gencode newer than runtime):
+  the build hook generated stubs with whatever `grpcio-tools` was newest, independent of the
+  protobuf runtime in the environment
+- Editable installs (`pip install -e .`) shadowing `livekit.agents`: the empty
+  `livekit/__init__.py` and `livekit/plugins/__init__.py` turned the shared namespace into a
+  regular package; they are removed
+
+### Changed
+- gRPC stubs now come from the `techmo-asr-api` package (1.1.4) instead of being generated
+  from `proto/` at build time; `hatch_build.py`, `proto/` and the `grpcio-tools` build
+  dependency are removed, and `--no-build-isolation` is no longer needed
+- Dependencies: `protobuf>=5.29.0` and `grpcio>=1.70.0` (required by the `techmo-asr-api`
+  stubs), `livekit-agents<2.0`
+- CI now runs on pushes and pull requests to `master` (it targeted a non-existent `main`)
+- README requirements aligned with the package metadata
+- Lint, format and mypy issues fixed so the CI passes; mypy no longer pins `python_version`
+
 ## [0.2.0] - 2026-09-30
 
 ### Added
diff --git a/Makefile b/Makefile
index e166fd7..b8ee3d0 100644
--- a/Makefile
+++ b/Makefile
@@ -1,14 +1,6 @@
-.PHONY: proto install install-dev lint format typecheck test clean
+.PHONY: install install-dev lint format typecheck test clean
 
-PROTO_DIR := proto
-OUTPUT_DIR := livekit/plugins/techmo/_proto
-
-# Generate gRPC/protobuf Python stubs from .proto files
-proto:
-	pip install grpcio-tools --quiet
-	python hatch_build.py
-
-# Install the package in editable/dev mode (also generates protos)
+# Install the package in editable/dev mode
 install-dev:
 	pip install -e ".[dev]"
 
@@ -28,7 +20,6 @@ test:
 	pytest tests/ -v
 
 clean:
-	rm -rf $(OUTPUT_DIR)/techmo
 	rm -rf build dist *.egg-info
 	find . -type d -name __pycache__ -exec rm -rf {} + 2>/dev/null || true
 	find . -name "*.pyc" -delete
diff --git a/README.md b/README.md
index dda7728..524eb0e 100644
--- a/README.md
+++ b/README.md
@@ -11,42 +11,34 @@
 - TLS and mutual-TLS support
 - Multiple language groups / model selection
 - MRCP timeout controls (no-input, recognition, speech-complete, speech-incomplete)
-- Compatible with LiveKit Agents >= 1.5
+- Compatible with LiveKit Agents 1.x
 
 ## Requirements
 
 - Python >= 3.10
-- `livekit-agents >= 1.5`
-- `grpcio >= 1.63`
-- `protobuf >= 5.0`
+- `livekit-agents >= 1.0, < 2.0` (tested with 1.5.1 and 1.8.3)
+- `grpcio >= 1.70`
+- `protobuf >= 5.29`
+- [`techmo-asr-api`](https://github.com/techmo-pl/asr-api-python) 1.1.4 (gRPC stubs; installed automatically)
 - Access to a running **Techmo ASR** gRPC server
 
 ## Installation
 
-### From source
-
 ```bash
-git clone https://github.com/techmo-pl/livekit-plugins-techmo
-cd livekit-plugins-techmo
-
-# Install build tools
-pip install grpcio-tools hatchling
-
-# Install the plugin (stubs are generated at build time)
-pip install --no-build-isolation .
+pip install "livekit-plugins-techmo @ git+https://github.com/techmo-pl/livekit-plugins-techmo.git@<version-tag>"
 ```
 
-> **Note:** The `--no-build-isolation` flag is required because the build hook generates
-> gRPC Python stubs from the `.proto` files in `proto/` at install time.
-> The stubs are placed in `livekit/plugins/techmo/_proto/`.
-
-To regenerate stubs manually after changing `.proto` files:
+or from a clone:
 
 ```bash
-pip install grpcio-tools
-python hatch_build.py
+git clone https://github.com/techmo-pl/livekit-plugins-techmo
+cd livekit-plugins-techmo
+pip install .
 ```
 
+The gRPC stubs come from the [`techmo-asr-api`](https://github.com/techmo-pl/asr-api-python)
+package (a dependency), so nothing is generated at install time.
+
 ## Quick Start
 
 ```python
@@ -166,9 +158,6 @@ LIVEKIT_LOG_LEVEL=DEBUG python my_agent.py dev
 ## Development
 
 ```bash
-# Generate gRPC stubs
-python hatch_build.py
-
 # Run linter
 ruff check .
 
@@ -184,8 +173,8 @@ TECHMO_ASR_ADDRESS=localhost:5555 pytest tests/test_integration.py -v
 
 ## API Version
 
-This plugin uses the **Techmo ASR v1p1** gRPC API. The `.proto` definition is located at
-`proto/techmo/asr/api/v1p1/asr.proto`.
+This plugin uses the **Techmo ASR v1p1** gRPC API (`asr_api.v1p1` from
+[`techmo-asr-api`](https://github.com/techmo-pl/asr-api-python), which also holds the `.proto` definitions).
 
 ## License
 
diff --git a/examples/basic_stt.py b/examples/basic_stt.py
index 1e0810e..473dfe5 100644
--- a/examples/basic_stt.py
+++ b/examples/basic_stt.py
@@ -20,9 +20,9 @@
     JobContext,
     WorkerOptions,
     cli,
-    llm,
 )
 from livekit.agents.pipeline import VoicePipelineAgent
+
 from livekit.plugins.techmo import STT
 
 logger = logging.getLogger("techmo-example")
@@ -35,7 +35,7 @@ async def entrypoint(ctx: JobContext) -> None:
     # service_address can also be read from TECHMO_ASR_ADDRESS env var
     stt_engine = STT(
         service_address=os.environ.get("TECHMO_ASR_ADDRESS", "localhost:5555"),
-        language_group="pl",       # Polish language group — adjust as needed
+        language_group="pl",  # Polish language group — adjust as needed
         interim_results=True,
         sample_rate=16000,
     )
diff --git a/examples/streaming_stt_standalone.py b/examples/streaming_stt_standalone.py
index 586f0b6..9257f45 100644
--- a/examples/streaming_stt_standalone.py
+++ b/examples/streaming_stt_standalone.py
@@ -12,7 +12,6 @@
 import asyncio
 import logging
 import os
-import struct
 import sys
 import wave
 
@@ -34,11 +33,10 @@ def read_wav_pcm(path: str) -> tuple[bytes, int]:
 async def transcribe_file(wav_path: str) -> None:
     try:
         from livekit import rtc
+
         from livekit.plugins.techmo.stt import STT
     except ImportError as e:
-        raise SystemExit(
-            "Install dependencies first: pip install livekit-plugins-techmo livekit-rtc"
-        ) from e
+        raise SystemExit("Install dependencies first: pip install livekit-plugins-techmo livekit-rtc") from e
 
     address = os.environ.get("TECHMO_ASR_ADDRESS", "localhost:5555")
     pcm, sample_rate = read_wav_pcm(wav_path)
diff --git a/hatch_build.py b/hatch_build.py
deleted file mode 100644
index 2cea38e..0000000
--- a/hatch_build.py
+++ /dev/null
@@ -1,113 +0,0 @@
-"""Hatchling build hook: generates gRPC/protobuf stubs from .proto files."""
-
-from __future__ import annotations
-
-import os
-import sys
-from pathlib import Path
-
-from hatchling.builders.hooks.plugin.interface import BuildHookInterface
-
-
-PROTO_DIR = Path(__file__).parent / "proto"
-OUTPUT_DIR = Path(__file__).parent / "livekit" / "plugins" / "techmo" / "_proto"
-
-PROTO_FILES = [
-    "techmo/api/status.proto",
-    "techmo/asr/api/v1p1/asr.proto",
-]
-
-
-class CustomBuildHook(BuildHookInterface):
-    PLUGIN_NAME = "custom"
-
-    def initialize(self, version: str, build_data: dict) -> None:
-        _generate_protos()
-
-
-def _generate_protos() -> None:
-    try:
-        from grpc_tools import protoc
-    except ImportError:
-        print(
-            "WARNING: grpcio-tools not found. Skipping proto generation. "
-            "Run 'pip install grpcio-tools && python -m grpc_tools.protoc ...' manually.",
-            file=sys.stderr,
-        )
-        return
-
-    OUTPUT_DIR.mkdir(parents=True, exist_ok=True)
-
-    # Write __init__.py so the package is importable
-    init_file = OUTPUT_DIR / "__init__.py"
-    if not init_file.exists():
-        init_file.write_text("# Auto-generated gRPC stubs package\n")
-
-    # Collect include paths: our proto dir + grpc_tools bundled protos
-    import grpc_tools
-
-    grpc_tools_proto_dir = Path(grpc_tools.__file__).parent / "_proto"
-    include_paths = [str(PROTO_DIR), str(grpc_tools_proto_dir)]
-    include_args = [f"-I{p}" for p in include_paths]
-
-    for proto_file in PROTO_FILES:
-        args = [
-            "grpc_tools.protoc",
-            *include_args,
-            f"--python_out={OUTPUT_DIR}",
-            f"--grpc_python_out={OUTPUT_DIR}",
-            f"--pyi_out={OUTPUT_DIR}",
-            proto_file,
-        ]
-        ret = protoc.main(args)
-        if ret != 0:
-            raise RuntimeError(f"protoc failed for {proto_file} (exit code {ret})")
-
-    # Ensure every subdirectory under OUTPUT_DIR is a proper Python package
-    _ensure_init_files(OUTPUT_DIR)
-
-    # Fix absolute imports in all generated files (protoc generates absolute imports)
-    _fix_proto_imports(OUTPUT_DIR)
-
-    print(f"gRPC stubs generated in {OUTPUT_DIR}")
-
-
-def _ensure_init_files(output_dir: Path) -> None:
-    """Create __init__.py in every subdirectory so they are importable packages."""
-    for dirpath in output_dir.rglob("*"):
-        if dirpath.is_dir():
-            init = dirpath / "__init__.py"
-            if not init.exists():
-                init.write_text("")
-
-
-def _fix_proto_imports(output_dir: Path) -> None:
-    """Replace absolute package imports with relative ones in all generated pb2 files.
-
-    protoc generates absolute imports like ``from techmo.x import y_pb2`` which
-    break when the stubs are nested under ``_proto/``.  We compute the correct
-    relative prefix for each file based on its depth under output_dir.
-    """
-    for pb_file in output_dir.glob("**/*.py"):
-        content = pb_file.read_text()
-        if "from techmo." not in content:
-            continue
-
-        # Compute how many package levels separate this file from output_dir.
-        # Each level = one dot in the relative import prefix.
-        depth = len(pb_file.relative_to(output_dir).parts) - 1  # -1 for the file itself
-        prefix = "." * depth  # e.g. depth=4 → "...." for _proto/techmo/asr/api/v1p1/
-
-        # Replace every absolute "from techmo.<sub> import" with a relative one.
-        # e.g. depth=4, "from techmo.api import" → "from ....api import"
-        import re
-        content = re.sub(
-            r"from techmo((?:\.\w+)*) import",
-            lambda m: f"from {prefix}{m.group(1).lstrip('.')} import",
-            content,
-        )
-        pb_file.write_text(content)
-
-
-if __name__ == "__main__":
-    _generate_protos()
diff --git a/livekit/__init__.py b/livekit/__init__.py
deleted file mode 100644
index e69de29..0000000
diff --git a/livekit/plugins/__init__.py b/livekit/plugins/__init__.py
deleted file mode 100644
index e69de29..0000000
diff --git a/livekit/plugins/techmo/__init__.py b/livekit/plugins/techmo/__init__.py
index 59c068f..294de82 100644
--- a/livekit/plugins/techmo/__init__.py
+++ b/livekit/plugins/techmo/__init__.py
@@ -3,7 +3,7 @@
 from livekit.agents import Plugin
 
 from .log import logger
-from .stt import STT, STTOptions, SpeechStream
+from .stt import STT, SpeechStream, STTOptions
 from .version import __version__
 
 __all__ = [
diff --git a/livekit/plugins/techmo/_proto/__init__.py b/livekit/plugins/techmo/_proto/__init__.py
deleted file mode 100644
index 955987a..0000000
--- a/livekit/plugins/techmo/_proto/__init__.py
+++ /dev/null
@@ -1,55 +0,0 @@
-"""Auto-generated gRPC/protobuf stubs for Techmo ASR v1p1.
-
-This package is generated by the build system. Run:
-
-    python hatch_build.py
-
-or
-
-    make proto
-
-to regenerate from the .proto files in the proto/ directory.
-"""
-
-try:
-    # Protoc places generated files in a nested path matching the proto package.
-    # After generation into _proto/, the files will be at:
-    #   _proto/techmo/asr/api/v1p1/asr_pb2.py
-    #   _proto/techmo/asr/api/v1p1/asr_pb2_grpc.py
-    from .techmo.asr.api.v1p1.asr_pb2 import (  # type: ignore
-        Audio,
-        AudioConfig,
-        ResultConfig,
-        SpeechRecognitionConfig,
-        SpeechRecognitionResult,
-        StreamingConfig,
-        StreamingRecognizeRequest,
-        StreamingRecognizeRequestConfig,
-        StreamingRecognizeRequestControlMessage,
-        StreamingRecognizeRequestData,
-        StreamingRecognizeResponse,
-        StreamingRecognizeResult,
-    )
-    from .techmo.asr.api.v1p1.asr_pb2_grpc import AsrStub  # type: ignore
-
-    __all__ = [
-        "Audio",
-        "AudioConfig",
-        "AsrStub",
-        "ResultConfig",
-        "SpeechRecognitionConfig",
-        "SpeechRecognitionResult",
-        "StreamingConfig",
-        "StreamingRecognizeRequest",
-        "StreamingRecognizeRequestConfig",
-        "StreamingRecognizeRequestControlMessage",
-        "StreamingRecognizeRequestData",
-        "StreamingRecognizeResponse",
-        "StreamingRecognizeResult",
-    ]
-
-except ImportError as exc:
-    raise ImportError(
-        "Techmo ASR gRPC stubs have not been generated yet.\n"
-        "Run 'python hatch_build.py' or 'make proto' from the project root."
-    ) from exc
diff --git a/livekit/plugins/techmo/stt.py b/livekit/plugins/techmo/stt.py
index 55251cf..7c05d95 100644
--- a/livekit/plugins/techmo/stt.py
+++ b/livekit/plugins/techmo/stt.py
@@ -3,29 +3,25 @@
 from __future__ import annotations
 
 import asyncio
-import dataclasses
 import os
-import time
+from collections.abc import AsyncIterator, Callable
 from dataclasses import dataclass
-from typing import AsyncIterator
+from typing import Any
 
 import grpc
 import grpc.aio
-
 from livekit import rtc
 from livekit.agents import stt, utils
 from livekit.agents.types import DEFAULT_API_CONNECT_OPTIONS, NOT_GIVEN, APIConnectOptions, NotGivenOr
 
 from .log import logger
-from .version import __version__
 
 try:
-    from . import _proto  # type: ignore[attr-defined]
+    from asr_api import v1p1 as _proto
 except ImportError as e:
     raise ImportError(
-        "Techmo ASR gRPC stubs not found. "
-        "Run 'pip install grpcio-tools && python hatch_build.py' to generate them, "
-        "or reinstall the package: 'pip install livekit-plugins-techmo'"
+        "Techmo ASR gRPC stubs (techmo-asr-api) not found. Reinstall the plugin, or install them with: "
+        "pip install 'techmo-asr-api @ git+https://github.com/techmo-pl/asr-api-python.git@1.1.4'"
     ) from e
 
 
@@ -149,8 +145,7 @@ def __init__(
             service_address = os.environ.get("TECHMO_ASR_ADDRESS")
         if not service_address:
             raise ValueError(
-                "service_address is required. "
-                "Pass it directly or set the TECHMO_ASR_ADDRESS environment variable."
+                "service_address is required. Pass it directly or set the TECHMO_ASR_ADDRESS environment variable."
             )
 
         self._opts = STTOptions(
@@ -185,7 +180,7 @@ def _make_channel(self) -> grpc.aio.Channel:
             return grpc.aio.secure_channel(opts.service_address, credentials)
         return grpc.aio.insecure_channel(opts.service_address)
 
-    async def _recognize_impl(
+    async def _recognize_impl(  # type: ignore[override]
         self,
         buffer: utils.AudioBuffer,
         *,
@@ -193,9 +188,7 @@ async def _recognize_impl(
     ) -> stt.SpeechEvent:
         """Perform batch (non-streaming) recognition on a complete audio buffer."""
         config = _build_config(self._opts)
-        audio_bytes = b"".join(
-            frame.data.tobytes() for frame in buffer
-        )
+        audio_bytes = b"".join(frame.data.tobytes() for frame in buffer)  # type: ignore[union-attr]
 
         async with self._make_channel() as channel:
             stub = _proto.AsrStub(channel)
@@ -226,7 +219,7 @@ async def _requests() -> AsyncIterator[_proto.StreamingRecognizeRequest]:  # typ
 
         return stt.SpeechEvent(
             type=stt.SpeechEventType.FINAL_TRANSCRIPT,
-            alternatives=alternatives or [stt.SpeechData(language="", text="")],
+            alternatives=alternatives or [stt.SpeechData(language="", text="")],  # type: ignore[arg-type]
             recognition_usage=stt.RecognitionUsage(audio_duration=duration),
         )
 
@@ -236,7 +229,7 @@ def stream(
         language: NotGivenOr[str] = NOT_GIVEN,
         conn_options: APIConnectOptions = DEFAULT_API_CONNECT_OPTIONS,
         **kwargs: object,
-    ) -> "SpeechStream":
+    ) -> SpeechStream:
         """Return an async streaming speech recognition context."""
         return SpeechStream(self, self._opts, conn_options=conn_options)
 
@@ -274,9 +267,7 @@ async def _request_generator() -> AsyncIterator[_proto.StreamingRecognizeRequest
                         )
                     )
 
-            recv_task = asyncio.ensure_future(
-                self._recv_loop(stub, _request_generator, send_queue)
-            )
+            recv_task = asyncio.ensure_future(self._recv_loop(stub, _request_generator, send_queue))
 
             resampler: rtc.AudioResampler | None = None
 
@@ -303,12 +294,12 @@ async def _request_generator() -> AsyncIterator[_proto.StreamingRecognizeRequest
 
     async def _recv_loop(
         self,
-        stub: object,
-        request_gen_factory: object,
+        stub: Any,
+        request_gen_factory: Callable[[], AsyncIterator[Any]],
         send_queue: asyncio.Queue[bytes | None],
     ) -> None:
         try:
-            async for response in stub.StreamingRecognize(  # type: ignore[union-attr]
+            async for response in stub.StreamingRecognize(
                 request_gen_factory(),
                 timeout=self._opts.grpc_timeout,
                 metadata=_build_metadata(self._opts),
@@ -408,7 +399,7 @@ def _build_metadata(opts: STTOptions) -> tuple[tuple[str, str], ...] | None:
     return ((_SESSION_ID_METADATA_KEY, opts.session_id),)
 
 
-def _build_config(opts: STTOptions) -> "_proto.StreamingRecognizeRequestConfig":  # type: ignore[name-defined]
+def _build_config(opts: STTOptions) -> _proto.StreamingRecognizeRequestConfig:  # type: ignore[name-defined]
     """Build the initial gRPC config message from plugin options."""
     speech_cfg_kwargs: dict[str, object] = {
         "enable_speech_recognition": True,
@@ -460,7 +451,7 @@ def _parse_alternatives(result: object) -> list[stt.SpeechData]:
         if not text:
             continue
         confidence = alt.confidence if alt.HasField("confidence") else 0.0
-        alts.append(stt.SpeechData(language="", text=text, confidence=confidence))
+        alts.append(stt.SpeechData(language="", text=text, confidence=confidence))  # type: ignore[arg-type]
 
     return alts
 
diff --git a/proto/google/rpc/status.proto b/proto/google/rpc/status.proto
deleted file mode 100644
index 923e169..0000000
--- a/proto/google/rpc/status.proto
+++ /dev/null
@@ -1,49 +0,0 @@
-// Copyright 2022 Google LLC
-//
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//     http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-
-syntax = "proto3";
-
-package google.rpc;
-
-import "google/protobuf/any.proto";
-
-option cc_enable_arenas = true;
-option go_package = "google.golang.org/genproto/googleapis/rpc/status;status";
-option java_multiple_files = true;
-option java_outer_classname = "StatusProto";
-option java_package = "com.google.rpc";
-option objc_class_prefix = "RPC";
-
-// The `Status` type defines a logical error model that is suitable for
-// different programming environments, including REST APIs and RPC APIs. It is
-// used by [gRPC](https://github.com/grpc). Each `Status` message contains
-// three pieces of data: error code, error message, and error details.
-//
-// You can find out more about this error model and how to work with it in the
-// [API Design Guide](https://cloud.google.com/apis/design/errors).
-message Status {
-  // The status code, which should be an enum value of
-  // [google.rpc.Code][google.rpc.Code].
-  int32 code = 1;
-
-  // A developer-facing error message, which should be in English. Any
-  // user-facing error message should be localized and sent in the
-  // [google.rpc.Status.details][google.rpc.Status.details] field, or localized
-  // by the client.
-  string message = 2;
-
-  // A list of messages that carry the error details.  There is a common set of
-  // message types for APIs to use.
-  repeated google.protobuf.Any details = 3;
-}
diff --git a/proto/techmo/api/status.proto b/proto/techmo/api/status.proto
deleted file mode 100644
index 71d4ad2..0000000
--- a/proto/techmo/api/status.proto
+++ /dev/null
@@ -1,11 +0,0 @@
-syntax = "proto3";
-package techmo.api;
-
-// Status communicates the result of a recognition request.
-// A non-zero code indicates an error.
-message Status {
-  // Status code; 0 means success.
-  int32 code = 1;
-  // Human-readable description of the status.
-  string message = 2;
-}
diff --git a/proto/techmo/asr/api/v1p1/asr.proto b/proto/techmo/asr/api/v1p1/asr.proto
deleted file mode 100644
index f2af15a..0000000
--- a/proto/techmo/asr/api/v1p1/asr.proto
+++ /dev/null
@@ -1,481 +0,0 @@
-// Copyright 2023 Techmo sp. z o.o.
-
-syntax = "proto3";
-
-package techmo.asr.api.v1p1;
-
-import "google/protobuf/duration.proto";
-import "techmo/api/status.proto";
-
-
-// An automatic speech recognition (ASR) service providing a solution for
-// speech-to-text conversion extended by the assessment of additional speech
-// and speaker features.
-service Asr {
-  // Perform bidirectional streaming recognition.
-  rpc StreamingRecognize(stream StreamingRecognizeRequest)
-    returns (stream StreamingRecognizeResponse) {}
-}
-
-// A message streamed from the client through
-// the [`StreamingRecognize`](#StreamingRecognize) method.
-message StreamingRecognizeRequest {
-  oneof request_content {
-    // The immutable initial configuration of the request.
-    // Must be sent once in the request's first message.
-    StreamingRecognizeRequestConfig config = 1;
-
-    // The message controlling the processing flow of the request.
-    // May be sent multiple times except in the request's first message.
-    StreamingRecognizeRequestControlMessage control_message = 2;
-
-    // The data contents of the request itself.
-    // May be sent multiple times except in the request's first message.
-    StreamingRecognizeRequestData data = 3;
-  }
-}
-
-// A message holding configuration of
-// a [`StreamingRecognize`](#StreamingRecognize) request.
-message StreamingRecognizeRequestConfig {
-  // Part of the configuration for the request's audio content.
-  AudioConfig audio_config = 1;
-
-  // Part of the configuration for the request's result form.
-  ResultConfig result_config = 2;
-
-  // Part of the configuration for the request's processing flow.
-  StreamingConfig streaming_config = 3;
-
-  // Part of the configuration for speech recognition.
-  SpeechRecognitionConfig speech_recognition_config = 4;
-
-  // Part of the configuration for age recognition.
-  AgeRecognitionConfig age_recognition_config = 5;
-
-  // Part of the configuration for gender recognition.
-  GenderRecognitionConfig gender_recognition_config = 6;
-
-  // Part of the configuration for language recognition.
-  LanguageRecognitionConfig language_recognition_config = 7;
-}
-
-// Result configuration of
-// a [`StreamingRecognize`](#StreamingRecognize) request.
-message ResultConfig {
-  // The switch that toggles continuous recognition into single utterance mode.
-  // The service returns a final result for each end of utterance it detects in
-  // the audio, which may occur multiple times during a request.
-  // If enabled, the request terminates right after its first final result.
-  bool enable_single_utterance = 1;
-
-  // The switch that allows interim results.
-  // If enabled, results containing tentative hypotheses may be returned in
-  // addition to final ones.
-  // The service should silently ignore this field if it is unsupported.
-  bool enable_interim_results = 2;
-
-  // The switch to allow the service merging responses in the "hold response"
-  // state.
-  // If enabled and there is more than a single response held, the service does
-  // not return them in a batch. Instead, it tries to merge their results into
-  // a single response.
-  // The service should respond with the `INVALID_ARGUMENT` gRPC status code
-  // if the `recognition_alternatives_limit` field
-  // of the [`SpeechRecognitionConfig`](#SpeechRecognitionConfig) message is
-  // greater than 1.
-  // New in v1p1.
-  bool enable_held_responses_merging = 3;
-}
-
-// Streaming configuration of
-// a [`StreamingRecognize`](#StreamingRecognize) request.
-message StreamingConfig {
-  reserved 1; // bool enable_single_utterance = 1;
-
-  // The switch that enables manual control of the input timer.
-  // The timer imposes two constraints: one that finalizes recognition after
-  // a specified period unless speech is detected, and the other that limits
-  // the total time for an utterance. Manual control allows recognition to
-  // begin but delays enforcement of these constraints. The timer restarts
-  // after each detected end of utterance (each final result).
-  // If enabled, the timer does not start automatically. Instead, it can be
-  // initiated by sending
-  // a [`StreamingRecognizeRequestControlMessage`](#StreamingRecognizeRequestControlMessage)
-  // with the `start_input_timer` field set to `true` as needed. This should
-  // occur after the beginning of the request and be repeated after each final
-  // result.
-  bool enable_manual_input_timer = 2;
-
-  // The switch to automatically set the service in the "hold response" state
-  // at the beginning of the request and after each final result.
-  // The "hold response" state means that the internal recognition process
-  // continues, but results are kept, not returned. When needed, the state can
-  // be toggled into the "give response" state by sending
-  // the [`StreamingRecognizeRequestControlMessage`](#StreamingRecognizeRequestControlMessage)
-  // message with the `give_response` field set to `true`.
-  // In the "give response" state the service responds as soon as it is ready.
-  // Any held responses may be returned in a batch or as a single merged
-  // response, provided that the `enable_held_responses_merging` field
-  // of the [`ResultConfig`](#ResultConfig) message is set to `true`.
-  // New in v1p1.
-  bool enable_auto_hold_response = 3;
-}
-
-// Audio configuration of
-// a [`StreamingRecognize`](#StreamingRecognize) request.
-message AudioConfig {
-  // The possible audio encodings.
-  enum AudioEncoding {
-    // Unspecified audio encoding.
-    UNSPECIFIED = 0;
-
-    // Linear pulse-code modulation of uncompressed 16-bit signed little-endian
-    // samples.
-    LINEAR16 = 1;
-
-    // Free Lossless Audio Codec ([FLAC](https://wiki.xiph.org/FLAC)).
-    // The encoding requires only about half the bandwidth of `LINEAR16`.
-    // 16-bit and 24-bit samples. Not all fields in `STREAMINFO` are supported.
-    // When set, the service ignores the `sampling_rate_hz` field and detects
-    // the actual value from audio header instead.
-    FLAC = 2;
-
-    // Ogg Encapsulated Opus Audio Codec ([OggOpus](https://wiki.xiph.org/OggOpus)).
-    // When set, the service ignores the `sampling_rate_hz` field and detects
-    // the actual value from audio header instead.
-    OGG_OPUS = 6;
-
-    // MP3 (ISO/IEC 11172-3 and ISO/IEC 13818-3).
-    // Only constant bitrate.
-    // When set, the service ignores the `sampling_rate_hz` field and detects
-    // the actual value from audio header instead.
-    MP3 = 8;
-  }
-
-  // The encoding of the audio data sent in the request. Single channel (mono)
-  // audio is assumed.
-  // The service should respond with the `INVALID_ARGUMENT` gRPC status code
-  // if the encoding is `UNSPECIFIED`.
-  // The service should respond with the `FAILED_PRECONDITION` gRPC status code
-  // if the encoding is not supported.
-  AudioEncoding encoding = 1;
-
-  // The sampling rate of the audio data sent in the request.
-  // The service should silently ignore the field for encodings that are sent
-  // along wtih headers, and detect the value from them instead.
-  // The service should respond with the `INVALID_ARGUMENT` gRPC status code
-  // if the value is not greater than 0.
-  float sampling_rate_hz = 2;
-}
-
-// Configuration of age recognition.
-message AgeRecognitionConfig {
-  // The switch that enables age recognition for the request.
-  // If disabled or unspecified, the related results are excluded.
-  // The service responds with the `FAILED_PRECONDITION` gRPC status code
-  // if requested but not enabled.
-  bool enable_age_recognition = 1;
-}
-
-// Configuration of gender recognition.
-message GenderRecognitionConfig {
-  // The switch that enables gender recognition for the request.
-  // If disabled or unspecified, the related results are excluded.
-  // The service responds with the `FAILED_PRECONDITION` gRPC status code
-  // if requested but not enabled.
-  bool enable_gender_recognition = 1;
-}
-
-// Configuration of language recognition.
-message LanguageRecognitionConfig {
-  // The switch that enables language recognition for the request.
-  // If disabled or unspecified, the related results are excluded.
-  // The service responds with the `FAILED_PRECONDITION` gRPC status code
-  // if requested but not enabled.
-  bool enable_language_recognition = 1;
-}
-
-// Configuration for speech recognition.
-message SpeechRecognitionConfig {
-  // The switch that enables speech recognition for the request.
-  // If disabled or unspecified, the related results are excluded.
-  // The service responds with the `FAILED_PRECONDITION` gRPC status code
-  // if requested but not enabled.
-  bool enable_speech_recognition = 1;
-
-  // The maximum number of alternative transcriptions allowed to be included
-  // per response.
-  // The actual count received can be less than the specified value and may
-  // also be equal to 0. If unspecified or 0, one alternative is allowed to be
-  // returned too.
-  uint32 recognition_alternatives_limit = 2;
-
-  // The switch that enables additional time alignment of recognitions in word
-  // details.
-  // If enabled, the `words` field of
-  // a [`SpeechRecognitionAlternative`](#SpeechRecognitionAlternative) message
-  // includes a list of [`SpeechRecognitionWord`](#SpeechRecognitionWord)
-  // messages. Otherwise, it remains empty.
-  // The service responds with the `FAILED_PRECONDITION` gRPC status code
-  // if requested but not enabled.
-  bool enable_time_alignment = 3;
-
-  // The name of a language group of models to be used.
-  // If left unspecified, it backs to the service's default group.
-  // The service responds with the `NOT_FOUND` gRPC status code
-  // if the name is not registered.
-  string language_group_name = 4;
-
-  // The name of a model to be used.
-  // If left unspecified, it backs to the selected langugage group's default.
-  // The service responds with the `NOT_FOUND` gRPC status code
-  // if the name is not registered.
-  string model_name = 5;
-
-  // Deprecated.
-  // The additional advanced service-dependend configuration for its speech
-  // recognizer. It may be silently ignored.
-  map<string, string> config_fields = 6;
-}
-
-// A message controlling the processing flow of
-// a [`StreamingRecognize`](#StreamingRecognize) request.
-message StreamingRecognizeRequestControlMessage {
-  reserved 2;
-
-  oneof control_message_content
-  {
-    // The flag that starts the input timer on demand and resets after each final
-    // result. It is silently ignored if the manual input timer setting is
-    // disabled for the request.
-    bool start_input_timer = 1;
-
-    // The flag to allow the service to return a response.
-    // After receiving this message, the service remains in the "give response"
-    // state. Ignored when the service is already in the "give response" state.
-    // Mutually exclusive with the `hold_response` field.
-    // New in v1p1.
-    bool give_response = 3;
-
-    // The flag to forbid the service from returning a response.
-    // After receiving this message, the service remains in the "hold response"
-    // state. Ignored when the service is already in the "hold response" state.
-    // Mutually exclusive with the `give_response` field.
-    // New in v1p1.
-    bool hold_response = 4;
-  }
-}
-
-// A message that carries data contents of
-// a [`StreamingRecognizeRequest`](#StreamingRecognizeRequest) request.
-message StreamingRecognizeRequestData {
-  // Part of the audio to perform recognition on.
-  Audio audio = 1;
-}
-
-// Audio contents.
-message Audio {
-  oneof audio_content {
-    // The audio data bytes.
-    bytes bytes = 1;
-  }
-}
-
-// A message streamed from the service through
-// the [`StreamingRecognize`](#StreamingRecognize) method.
-message StreamingRecognizeResponse {
-  // The combined recognition results for another part of the audio.
-  StreamingRecognizeResult result = 1;
-
-  // The cumulative duration of the processed audio during the request,
-  // not necessarily matching the actual length of the sent audio, mandatorily
-  // updated with each final result.
-  google.protobuf.Duration processed_audio_duration = 2;
-}
-
-// Combined recognition result.
-message StreamingRecognizeResult {
-  // The recognition process status.
-  // It may communicate warnings. In case of an error hindering recognition,
-  // all other message fields should be left unset.
-  techmo.api.Status error = 1;
-
-  // The flag indicating whether the result is interim or final.
-  bool is_final = 2;
-
-  // The anticipated causes for the service to finalize a result.
-  enum ResultFinalizationCause {
-    // The cause is not specified.
-    UNSPECIFIED = 0;
-
-    // The speech recognition result is not empty and the end of utterance
-    // is detected.
-    SUCCESS = 1;
-
-    // The speech recognition result is empty after the duration to expect
-    // a result is reached.
-    NO_INPUT_TIMEOUT = 2;
-
-    // The speech recognition result is not empty after the utterance duration
-    // limit is reached. The returned speech recognition is incomplete and
-    // should be completed in the following result.
-    SUCCESS_MAXTIME = 3;
-
-    // Unused.
-    PARTIAL_MATCH = 4;
-
-    // The speech recognition result is empty after the utterance duration
-    // limit is reached.
-    NO_MATCH_MAXTIME = 5;
-  }
-
-  // The field indicating the cause of result finalization.
-  // For interim results, the service should leave the field as `UNSPECIFIED`.
-  // For final results, the service must set the field to a value other than
-  // `UNSPECIFIED`.
-  ResultFinalizationCause result_finalization_cause = 3;
-
-  // The speech recognition result for another part of the processed audio,
-  // new with each final result, updates with each interim one.
-  // To obtain a complete result for all processed audio, for each final result
-  // received, a client should pick one of the result's recognition alternatives
-  // and buffer it on its own.
-  // It must be omitted if speech recognition is disabled.
-  SpeechRecognitionResult speech_recognition_result = 4;
-
-  // The current age recognition result for all processed audio,
-  // updated with each final result.
-  // It may be omitted in an interim result and must be omitted if age
-  // recognition is disabled.
-  AgeRecognitionResult age_recognition_result = 5;
-
-  // The current gender recognition result for all processed audio,
-  // updated with each final result.
-  // It may be omitted in an interim result and must be omitted if gender
-  // recognition is disabled.
-  GenderRecognitionResult gender_recognition_result = 6;
-
-  // The current language recognition result for all processed audio,
-  // updated with each final result.
-  // It may be omitted in an interim result and must be omitted if language
-  // recognition is disabled.
-  LanguageRecognitionResult language_recognition_result = 7;
-}
-
-// A result of age recognition.
-message AgeRecognitionResult {
-  // The recognition process status.
-  // It may communicate warnings. In case of an error hindering recognition,
-  // all other message fields should be left unset.
-  techmo.api.Status error = 1;
-
-  // The confidence-ordered list of alternative recognition hypotheses.
-  repeated AgeRecognitionAlternative recognition_alternatives = 2;
-}
-
-// An alternative hypothesis of age recognition.
-message AgeRecognitionAlternative {
-  // The assumed age of the person speaking in the audio, in years.
-  // For a reliable value, assure that there is only one person speaking in
-  // the audio.
-  uint32 age = 1;
-
-  // The confidence estimate, ranging from 0.0 to 1.0.
-  // Support for this feature is optional.
-  optional float confidence = 2;
-}
-
-// A result of gender recognition.
-message GenderRecognitionResult {
-  // The recognition process status.
-  // It may communicate warnings. In case of an error hindering recognition,
-  // all other message fields should be left unset.
-  techmo.api.Status error = 1;
-
-  // The confidence-ordered list of alternative recognition hypotheses.
-  repeated GenderRecognitionAlternative recognition_alternatives = 2;
-}
-
-// An alternative hypothesis of gender recognition.
-message GenderRecognitionAlternative {
-  // The assumed gender of the person speaking in the audio.
-  // For a reliable value, assure that there is only one person speaking in
-  // the audio.
-  string gender = 1;
-
-  // The confidence estimate, ranging from 0.0 to 1.0.
-  // Support for this feature is optional.
-  optional float confidence = 2;
-}
-
-// A result of language recognition.
-message LanguageRecognitionResult {
-  // The recognition process status.
-  // It may communicate warnings. In case of an error hindering recognition,
-  // all other message fields should be left unset.
-  techmo.api.Status error = 1;
-
-  // The confidence-ordered list of alternative recognition hypotheses.
-  repeated LanguageRecognitionAlternative recognition_alternatives = 2;
-}
-
-// An alternative hypothesis of language recognition.
-message LanguageRecognitionAlternative {
-  // The language spoken in the audio,
-  // a [BCP-47](https://www.rfc-editor.org/rfc/bcp/bcp47.txt) tag.
-  string language = 1;
-
-  // The confidence estimate, ranging from 0.0 to 1.0.
-  // Support for this feature is optional.
-  optional float confidence = 2;
-}
-
-// A result of speech recognition.
-message SpeechRecognitionResult {
-  // The recognition process status.
-  // It may communicate warnings. In case of an error hindering recognition,
-  // all other message fields should be left unset.
-  techmo.api.Status error = 1;
-
-  // The confidence-ordered list of alternative recognition hypotheses.
-  repeated SpeechRecognitionAlternative recognition_alternatives = 2;
-
-  // The actual name of the language group of the model,
-  // unrelated to the actual language spoken in the audio.
-  string language_group_name = 3;
-
-  // The actual name of the model used to obtain the result.
-  string model_name = 4;
-}
-
-// An alternative hypothesis of speech recognition.
-message SpeechRecognitionAlternative {
-  // The transcript of the audio.
-  string transcript = 1;
-
-  // The confidence estimate, ranging from 0.0 to 1.0.
-  // Support for this feature is optional.
-  optional float confidence = 2;
-
-  // The details of the transcript's words.
-  // Empty unless `enable_time_alignment` is `true` in the request's
-  // [`SpeechRecognitionConfig`](#SpeechRecognitionConfig).
-  repeated SpeechRecognitionWord words = 3;
-}
-
-// Details of a single word in speech recognition.
-message SpeechRecognitionWord {
-  // The transcript of the word itself.
-  string transcript = 1;
-
-  // The confidence estimate, ranging from 0.0 to 1.0.
-  // Support for this feature is optional.
-  optional float confidence = 2;
-
-  // The start time of the word relative to the beginning of the entire audio.
-  google.protobuf.Duration start_time = 3;
-
-  // The end time of the word relative to the beginning of the entire audio.
-  google.protobuf.Duration end_time = 4;
-}
diff --git a/pyproject.toml b/pyproject.toml
index e65fc2e..a252a3d 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -1,5 +1,5 @@
 [build-system]
-requires = ["hatchling", "grpcio-tools>=1.63.0"]
+requires = ["hatchling"]
 build-backend = "hatchling.build"
 
 [project]
@@ -22,9 +22,12 @@ classifiers = [
     "Topic :: Multimedia :: Sound/Audio :: Speech",
 ]
 dependencies = [
-    "livekit-agents>=1.0",
-    "grpcio>=1.63.0",
-    "protobuf>=5.0.0",
+    "livekit-agents>=1.0,<2.0",
+    # techmo-asr-api stubs embed GRPC_GENERATED_VERSION 1.70.0 (older grpcio raises on import)
+    "grpcio>=1.70.0",
+    # techmo-asr-api ships stubs generated with protobuf 5.29 (older runtimes fail to import them)
+    "protobuf>=5.29.0",
+    "techmo-asr-api @ git+https://github.com/techmo-pl/asr-api-python.git@1.1.4",
 ]
 
 [project.urls]
@@ -39,20 +42,17 @@ dev = [
     "pytest-asyncio>=0.23",
     "ruff>=0.1.0",
     "mypy>=1.0",
-    "grpcio-tools>=1.63.0",
 ]
 
-[tool.hatch.build.hooks.custom]
-path = "hatch_build.py"
+[tool.hatch.metadata]
+# techmo-asr-api is not on PyPI yet, so it is referenced by git URL
+allow-direct-references = true
 
 [tool.hatch.build.targets.wheel]
 packages = ["livekit"]
-artifacts = [
-    "livekit/plugins/techmo/_proto/techmo/",
-]
 
 [tool.hatch.build.targets.sdist]
-include = ["livekit", "proto", "hatch_build.py"]
+include = ["livekit"]
 
 [tool.ruff]
 line-length = 120
@@ -61,8 +61,13 @@ target-version = "py310"
 [tool.ruff.lint]
 select = ["E", "F", "I", "UP"]
 
+[tool.ruff.lint.isort]
+# livekit/ is a namespace package shared with livekit-agents and livekit-rtc
+known-third-party = ["livekit"]
+known-first-party = ["livekit.plugins.techmo"]
+
 [tool.mypy]
-python_version = "3.10"
+# no python_version pin: dependency stubs (e.g. numpy) use syntax of the running interpreter
 strict = false
 ignore_missing_imports = true
 
diff --git a/tests/test_integration.py b/tests/test_integration.py
index 94f812a..457cfb4 100644
--- a/tests/test_integration.py
+++ b/tests/test_integration.py
@@ -28,6 +28,7 @@ async def test_streaming_silence() -> None:
     """Send silence and expect either empty results or graceful handling."""
     try:
         from livekit import rtc
+
         from livekit.plugins.techmo.stt import STT
     except ImportError:
         pytest.skip("livekit-agents or gRPC stubs not installed")
diff --git a/tests/test_stt.py b/tests/test_stt.py
index 28ebfab..4ca1d7a 100644
--- a/tests/test_stt.py
+++ b/tests/test_stt.py
@@ -8,7 +8,7 @@
 
 import asyncio
 import struct
-from unittest.mock import AsyncMock, MagicMock, patch
+from unittest.mock import MagicMock
 
 import pytest
 
@@ -63,13 +63,13 @@ def test_sttoptions_custom() -> None:
 def test_stt_requires_address(monkeypatch: pytest.MonkeyPatch) -> None:
     monkeypatch.delenv("TECHMO_ASR_ADDRESS", raising=False)
     with pytest.raises(ValueError, match="service_address"):
-        # Importing STT triggers proto import; skip if stubs not generated
+        # Importing STT triggers the techmo-asr-api import; skip if it is missing
         try:
             from livekit.plugins.techmo.stt import STT
 
             STT()
         except ImportError:
-            pytest.skip("gRPC stubs not generated")
+            pytest.skip("gRPC stubs (techmo-asr-api) not installed")
 
 
 def test_stt_address_from_env(monkeypatch: pytest.MonkeyPatch) -> None:
@@ -80,12 +80,13 @@ def test_stt_address_from_env(monkeypatch: pytest.MonkeyPatch) -> None:
         stt = STT()
         assert stt._opts.service_address == "localhost:50051"
     except ImportError:
-        pytest.skip("gRPC stubs not generated")
+        pytest.skip("gRPC stubs (techmo-asr-api) not installed")
 
 
 def test_stt_capabilities() -> None:
     try:
-        from livekit.agents import stt as lk_stt
+        from livekit.agents import stt as lk_stt  # noqa: F401  (availability check)
+
         from livekit.plugins.techmo.stt import STT
 
         instance = STT(service_address="localhost:50051")
@@ -110,7 +111,7 @@ def test_build_config_basic() -> None:
         assert cfg.result_config.enable_interim_results is True
         assert cfg.speech_recognition_config.enable_speech_recognition is True
     except ImportError:
-        pytest.skip("gRPC stubs not generated")
+        pytest.skip("gRPC stubs (techmo-asr-api) not installed")
 
 
 def test_build_config_with_model() -> None:
@@ -128,7 +129,7 @@ def test_build_config_with_model() -> None:
         assert cfg.speech_recognition_config.model_name == "asr-pl-v2"
         assert cfg.speech_recognition_config.recognition_alternatives_limit == 3
     except ImportError:
-        pytest.skip("gRPC stubs not generated")
+        pytest.skip("gRPC stubs (techmo-asr-api) not installed")
 
 
 # ---------------------------------------------------------------------------
@@ -145,7 +146,7 @@ def test_proto_duration_to_seconds() -> None:
         dur.nanos = 500_000_000
         assert _proto_duration_to_seconds(dur) == pytest.approx(2.5)
     except ImportError:
-        pytest.skip("gRPC stubs not generated")
+        pytest.skip("gRPC stubs (techmo-asr-api) not installed")
 
 
 # ---------------------------------------------------------------------------