diff --git a/authors/assets/goodguypeci-design.png b/authors/assets/goodguypeci-design.png
new file mode 100644
index 00000000..ec7ae083
Binary files /dev/null and b/authors/assets/goodguypeci-design.png differ
diff --git a/authors/petar-ivanov.md b/authors/petar-ivanov.md
new file mode 100644
index 00000000..fb3e2faf
--- /dev/null
+++ b/authors/petar-ivanov.md
@@ -0,0 +1,10 @@
+Author: Petar Ivanov
+Title: Contributor
+Description: Petar Ivanov contributed this guide to running Sapat with Azure AI Speech in a Daytona sandbox. The guide covers setup, a first transcription, and checks for common configuration and output problems.
+Author Image: ![GitHub account avatar for goodguypeci-design](./assets/goodguypeci-design.png)
+Author LinkedIn: Not provided
+Author Twitter: Not provided
+Company Name: Not provided
+Company Description: Not provided
+Company Logo Dark: Not applicable
+Company Logo White: Not applicable
diff --git a/definitions/20260908_definition_speech_to_text_api.md b/definitions/20260908_definition_speech_to_text_api.md
new file mode 100644
index 00000000..32825cb7
--- /dev/null
+++ b/definitions/20260908_definition_speech_to_text_api.md
@@ -0,0 +1,23 @@
+---
+title: 'Speech-to-text API'
+description: 'A service interface that accepts recorded speech and returns recognized text.'
+date: 2026-09-08
+author: 'GoodGuyPeci Design'
+---
+
+# Speech-to-text API
+
+## Definition
+
+A speech-to-text API is a programmatic service that receives audio containing
+spoken language and returns a text transcription. Applications usually send an
+audio file or stream together with settings such as the spoken language and the
+desired response format.
+
+## Context and Usage
+
+Speech-to-text APIs are used for meeting notes, subtitles, searchable media,
+voice interfaces, and call analysis. A client must prepare audio in a format the
+provider accepts, authenticate the request, handle size or duration limits, and
+store the returned transcript. Tools such as Sapat hide much of that provider
+specific work behind one command-line interface.
diff --git a/guides/20260908_azure_ai_speech_transcription_with_sapat.md b/guides/20260908_azure_ai_speech_transcription_with_sapat.md
new file mode 100644
index 00000000..8edd520d
--- /dev/null
+++ b/guides/20260908_azure_ai_speech_transcription_with_sapat.md
@@ -0,0 +1,349 @@
+---
+title: 'Azure AI Speech Transcription with Sapat'
+description:
+  'Run Sapat in a Daytona sandbox and transcribe audio through the Azure AI
+  Speech short-audio REST API.'
+date: 2026-09-08
+author: 'Petar Ivanov'
+tags: ['Daytona', 'Sapat', 'Azure AI Speech', 'speech-to-text']
+---
+
+# Azure AI Speech Transcription with Sapat
+
+Sapat gives one command-line interface to several transcription services. That
+is useful when an application needs to change providers without rewriting its
+whole audio pipeline. The project already supports Azure OpenAI, but Azure has
+another product with a similar name: Azure AI Speech. Azure OpenAI expects a
+deployed Whisper model; Azure AI Speech uses a Speech resource and its own
+speech-to-text endpoint.
+
+This guide uses the `azure_speech` provider proposed in
+[Sapat pull request #80](https://github.com/nibzard/sapat/pull/80). You will put
+Sapat in a fresh Daytona sandbox, configure an Azure Speech resource, run a
+transcription, and check the output. The sandbox keeps the setup away from your
+main machine, which makes it easier to repeat the same work or discard it when
+you are finished.
+
+![Daytona sandbox runs Sapat, which sends converted audio to Azure AI Speech and saves the returned text as sample.txt in the sandbox](assets/20260908_azure_ai_speech_sapat_daytona_img1.png)
+
+## TL;DR
+
+- Create a Daytona sandbox and clone the Sapat feature branch into it.
+- Install Python dependencies and FFmpeg inside the sandbox.
+- Put the Azure Speech key and resource endpoint in an ignored `.env` file.
+- Run `sapat sample.mp4 --provider azure_speech --language en-US`.
+- Confirm that `sample.txt` contains the recognized speech.
+
+## What you need
+
+Have these ready before opening a terminal:
+
+| Item | Why it is needed |
+| --- | --- |
+| Daytona account and API key | Creates and controls the isolated sandbox. |
+| Daytona CLI | Runs sandbox commands from your computer. |
+| Azure subscription and Speech resource | Supplies the transcription service, key, and endpoint. |
+| A copy of a short MP4 with speech | Gives you a known input for the first test. |
+
+Use Bash for the commands with `\` line continuations, including when running
+them from Windows through WSL or Git Bash. The installer command below is the
+one exception: run it in PowerShell. Commands after `daytona ssh` run inside the
+Linux sandbox.
+
+Microsoft limits a direct short-audio request to 60 seconds. The provider asks
+Sapat for 16 kHz mono WAV audio and uses a conservative 1.8 MB chunk size, which
+keeps normal PCM chunks close to that limit. Longer videos can still be used:
+Sapat converts them with FFmpeg and splits the resulting audio before sending
+the pieces one at a time.
+
+## Step 1: Install and authenticate the Daytona CLI
+
+Daytona now calls its isolated environments _sandboxes_. If you have an older
+Daytona installation, check its version before following commands from an older
+workspace tutorial:
+
+```bash
+daytona version
+```
+
+On macOS, install the current CLI with Homebrew:
+
+```bash
+brew install daytonaio/cli/daytona
+```
+
+On Windows, run the official installer from PowerShell:
+
+```powershell
+powershell -Command "irm https://get.daytona.io/windows | iex"
+```
+
+Get an API key from the Daytona dashboard and log in:
+
+```bash
+daytona login --api-key=YOUR_DAYTONA_API_KEY
+```
+
+Do not copy the real key into documentation, a commit, or a shared terminal
+recording. Verify the connection with `daytona list`. An empty list is fine; it
+means authentication works and you have no sandboxes yet.
+
+## Step 2: Create a clean Sapat sandbox
+
+Create a named sandbox in the European region:
+
+```bash
+daytona create --name sapat-azure-speech --target eu
+```
+
+The name makes later commands easier to read. Confirm that Daytona started it:
+
+```bash
+daytona info sapat-azure-speech
+```
+
+Use `daytona exec` to install the system packages Sapat needs. `git` downloads
+the source; `ffmpeg` and `ffprobe` convert and inspect the media; the Python venv
+package creates an isolated package environment inside the sandbox.
+
+```bash
+daytona exec sapat-azure-speech -- bash -lc \
+  "sudo apt-get update && sudo apt-get install -y git ffmpeg python3-venv"
+```
+
+Next, clone the branch from the open implementation PR and install Sapat:
+
+```bash
+daytona exec sapat-azure-speech -- bash -lc \
+  "git clone --branch feat/azure-speech-provider \
+  https://github.com/goodguypeci-design/sapat.git workspace/sapat && \
+  cd workspace/sapat && python3 -m venv .venv && \
+  . .venv/bin/activate && python -m pip install --upgrade pip && \
+  python -m pip install -e ."
+```
+
+The fork and branch are needed while PR #80 is under review. After it is merged,
+clone `https://github.com/nibzard/sapat.git` without the `--branch` argument.
+
+Check the installation without contacting Azure:
+
+```bash
+daytona exec sapat-azure-speech -- bash -lc \
+  "cd workspace/sapat && . .venv/bin/activate && \
+  ffmpeg -version | head -n 1 && sapat --version"
+```
+
+## Step 3: Find the correct Azure values
+
+In the Azure portal, create or open an Azure AI Speech resource. On its **Keys
+and Endpoint** page, copy one resource key and the endpoint. A current endpoint
+normally looks like this:
+
+```text
+https://your-resource-name.cognitiveservices.azure.com
+```
+
+The endpoint matters because Microsoft builds the short-audio URL from the
+resource name. A region such as `westeurope` can be used as a fallback for an
+older or non-custom-domain setup, but it is not the preferred value for a new
+resource.
+
+Connect to the sandbox and move into the cloned repository:
+
+```bash
+daytona ssh sapat-azure-speech
+cd workspace/sapat
+```
+
+Create the local settings file:
+
+```bash
+cp .env.example .env
+chmod 600 .env
+```
+
+Open `.env` in your terminal editor and set these values:
+
+```dotenv
+AZURE_SPEECH_KEY=your_real_speech_resource_key
+AZURE_SPEECH_ENDPOINT=https://your-resource-name.cognitiveservices.azure.com
+AZURE_SPEECH_RESPONSE_FORMAT=detailed
+AZURE_SPEECH_PROFANITY=masked
+```
+
+Sapat's `.gitignore` excludes `.env`, MP4, MP3, WAV, and TXT files.
+Still run `git status --short` before every push. The command should not show
+`.env` or your sample media.
+
+For a quick private experiment, the ignored file is the shortest route. For a
+long-lived or shared environment, use
+[Daytona Secrets](https://www.daytona.io/docs/en/secrets/) instead. A Daytona
+Secret keeps the real key outside the sandbox and can restrict substitution to
+your `*.cognitiveservices.azure.com` host. Azure accepts this key in the
+`Ocp-Apim-Subscription-Key` HTTPS header, so it fits Daytona's header-based
+secret substitution model.
+
+## Step 4: Put a sample in the sandbox
+
+If the media is already in a repository the sandbox can clone, place it under
+`workspace/sapat` and skip this step. Keep private recordings out of public Git
+history.
+
+For a local file, the Daytona Python SDK can upload it directly. Open a second
+terminal on your own computer, leaving the SSH session open. Install the SDK
+with `python -m pip install daytona`. Set the
+[`DAYTONA_API_KEY` environment variable](../definitions/20241126_definition_environment_variables.md)
+in that terminal to the same key used for the CLI. The SDK needs its own environment
+configuration; a successful CLI login does not configure this Python process.
+
+Save the following as `upload_sample.py` alongside `sample.mp4`, then run
+`python upload_sample.py` from that directory:
+
+```python
+from pathlib import Path
+
+from daytona import Daytona
+
+daytona = Daytona()
+sandbox = daytona.get("sapat-azure-speech")
+sandbox.fs.upload_file(
+    Path("sample.mp4").read_bytes(),
+    "workspace/sapat/sample.mp4",
+)
+```
+
+The upload goes to the sandbox file system and does not add the recording to
+Git. For the first test, choose a clear 10-to-20-second clip whose words you
+already know. Use a copy named `sample.mp4`, with no existing `sample.wav` or
+`sample.txt` beside it. This branch reuses and then removes a same-named WAV
+file, and overwrites a same-named transcript. Do not use your only copy of a
+recording as the test input.
+
+## Step 5: Run the transcription
+
+Return to the SSH session, which is still in `workspace/sapat` from Step 3.
+Activate the virtual environment and run Sapat:
+
+```bash
+. .venv/bin/activate
+sapat sample.mp4 --provider azure_speech --language en-US
+```
+
+For Bulgarian audio, use `--language bg-BG`. Short aliases such as `en` and
+`bg` are also normalized by the provider, but an explicit locale is clearer in
+reproducible commands.
+
+Sapat first invokes FFmpeg to produce the 16 kHz, mono, 16-bit WAV format Azure
+expects. It then sends the bytes to the resource endpoint with the language and
+response format as query parameters. A successful response becomes
+`sample.txt`, and the temporary converted WAV file is removed.
+
+Read the output and compare it with the words in your clip. A created text file
+alone does not tell you whether the transcription is accurate. Also check that
+no temporary media was staged for Git:
+
+```bash
+cat sample.txt
+git status --short
+```
+
+You can also confirm that Sapat discovered the new provider before making a
+request:
+
+```bash
+python -c "from sapat.providers import get_provider_choices; print(get_provider_choices())"
+```
+
+The printed list should contain `azure_speech`. If it does not, check that `.env`
+is in the Sapat repository root and contains both the key and either the endpoint
+or region.
+
+## Step 6: Verify the provider without spending an API call
+
+The implementation includes mocked request tests. They inspect the URL, headers,
+language normalization, detailed-result parsing, `NoMatch`, and HTTP error
+messages without sending real audio to Azure:
+
+```bash
+python -m pip install -e ".[dev]"
+python -m pytest tests/providers/test_group_e.py tests/test_registry.py \
+  tests/test_cli.py tests/test_base.py -q
+```
+
+These tests check client behavior against simulated responses. They do not
+verify Azure credentials, service availability, or recognition accuracy. A
+successful request with your sample is a separate check that the key, endpoint,
+account, and media work together; listening and comparing the words checks the
+transcript itself.
+
+## Common issues and troubleshooting
+
+### `Provider 'azure_speech' is not available`
+
+Provider discovery only exposes services whose required configuration is
+present. Confirm the `.env` filename, its directory, and the variable names.
+Restart the command after editing because the environment is read when Sapat is
+imported.
+
+### Azure returns `401 Unauthorized`
+
+The key and endpoint usually belong to different resources, or a regional URL
+does not match the key's region. Copy both values again from the same **Keys and
+Endpoint** page. Do not add the recognition path yourself when
+`AZURE_SPEECH_ENDPOINT` contains only the resource root; the provider appends it.
+
+### Azure returns `400 Bad Request`
+
+Use a supported locale such as `en-US` or `bg-BG`, and let Sapat perform the
+conversion. Microsoft accepts PCM WAV at 16 kHz mono or Ogg Opus for this API.
+A damaged file or unsupported language also produces a 4xx response.
+
+### The transcript is empty
+
+Azure uses `NoMatch` when it detects audio but cannot match words in the chosen
+language. Try the exact locale spoken in the clip, reduce background noise, and
+test with a short recording whose contents are known.
+
+### A long recording fails on one chunk
+
+Check that FFmpeg and `ffprobe` are installed, then try a shorter source. Inspect
+the output for `[Chunk N transcription failed]` markers: this branch can write
+a transcript even when individual chunks fail. Keep the original recording so
+you can retry those sections. If your application regularly processes long
+recordings, review Microsoft's fast and batch transcription options in the
+[Speech API documentation](https://learn.microsoft.com/en-us/azure/ai-services/speech-service/rest-speech-to-text-short).
+
+## Clean up when finished
+
+Leave the SSH session with `exit`. Daytona can stop, archive, or delete the
+sandbox according to how long you need the files. List it first so you target
+the correct environment:
+
+```bash
+daytona list
+daytona stop sapat-azure-speech
+```
+
+Delete it only when you no longer need the transcript or setup:
+
+```bash
+daytona delete sapat-azure-speech
+```
+
+## Conclusion
+
+The workflow has three parts to check when something goes wrong: Daytona runs
+the environment, Sapat converts the media and saves the result, and Azure AI
+Speech recognizes the words. Start with a short clip, compare the transcript
+with what you hear, then try a longer recording and inspect every failed-chunk
+marker before using the text elsewhere.
+
+## References
+
+- [Sapat repository](https://github.com/nibzard/sapat)
+- [Azure AI Speech provider PR](https://github.com/nibzard/sapat/pull/80)
+- [Microsoft short-audio REST API](https://learn.microsoft.com/en-us/azure/ai-services/speech-service/rest-speech-to-text-short)
+- [Daytona CLI reference](https://www.daytona.io/docs/en/tools/cli/)
+- [Daytona authentication](https://www.daytona.io/docs/en/api-keys/)
+- [Daytona Secrets](https://www.daytona.io/docs/en/secrets/)
+- [Daytona file-system operations](https://www.daytona.io/docs/file-system-operations/)
diff --git a/guides/assets/20260908_azure_ai_speech_sapat_daytona_img1.png b/guides/assets/20260908_azure_ai_speech_sapat_daytona_img1.png
new file mode 100644
index 00000000..77ebd667
Binary files /dev/null and b/guides/assets/20260908_azure_ai_speech_sapat_daytona_img1.png differ