saral 0.1.0__tar.gz

saral-0.1.0/.gitignore

@@ -0,0 +1,17 @@
+__pycache__/
+*.py[cod]
+*.egg-info/
+.venv/
+venv/
+build/
+dist/
+.pytest_cache/
+.ruff_cache/
+.coverage
+htmlcov/
+.DS_Store
+# Local runtime data never belongs in git
+.saral/
+evals/results/
+*.gguf
+*.spec

saral-0.1.0/LICENSE

@@ -0,0 +1,202 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   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.

saral-0.1.0/NOTICE

@@ -0,0 +1,10 @@
+Saral
+Copyright 2026 Harishwar Reddy Korpol and the Saral contributors.
+
+This product is licensed under the Apache License, Version 2.0 (see LICENSE).
+
+"Saral" and the Saral logo are trademarks of the project. The Apache License
+does not grant permission to use these marks; see TRADEMARK.md.
+
+Saral calls, but does not bundle, third-party models and tools. Each keeps its
+own license. See docs/THIRD_PARTY.md for the list and the license traps we avoid.

saral-0.1.0/PKG-INFO

@@ -0,0 +1,174 @@
+Metadata-Version: 2.5
+Name: saral
+Version: 0.1.0
+Summary: Saral: a private, voice-first AI helper that runs at home. Explains papers, checks for scams, in your own language.
+Project-URL: Homepage, https://github.com/Fullcountr/saral
+Project-URL: Documentation, https://github.com/Fullcountr/saral/tree/main/docs
+Project-URL: Issues, https://github.com/Fullcountr/saral/issues
+Project-URL: Changelog, https://github.com/Fullcountr/saral/releases
+Author: Harishwar Reddy Korpol
+License-Expression: Apache-2.0
+License-File: LICENSE
+License-File: NOTICE
+Keywords: accessibility,local-ai,multilingual,ocr,offline,privacy,scam-detection
+Classifier: Development Status :: 3 - Alpha
+Classifier: Intended Audience :: Developers
+Classifier: Intended Audience :: End Users/Desktop
+Classifier: Natural Language :: English
+Classifier: Natural Language :: Hindi
+Classifier: Natural Language :: Spanish
+Classifier: Operating System :: OS Independent
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Classifier: Topic :: Security
+Requires-Python: >=3.10
+Requires-Dist: fastapi>=0.110
+Requires-Dist: httpx>=0.27
+Requires-Dist: pillow>=10.0
+Requires-Dist: psutil>=5.9
+Requires-Dist: pydantic>=2.6
+Requires-Dist: pytesseract>=0.3.10
+Requires-Dist: python-multipart>=0.0.9
+Requires-Dist: qrcode>=7.4
+Requires-Dist: uvicorn>=0.29
+Provides-Extra: all
+Requires-Dist: faster-whisper>=1.0; extra == 'all'
+Provides-Extra: build
+Requires-Dist: pyinstaller>=6.6; extra == 'build'
+Provides-Extra: dev
+Requires-Dist: pytest>=8.0; extra == 'dev'
+Requires-Dist: ruff>=0.5; extra == 'dev'
+Provides-Extra: ocr
+Provides-Extra: speech
+Requires-Dist: faster-whisper>=1.0; extra == 'speech'
+Description-Content-Type: text/markdown
+
+# Saral
+
+**A private AI helper that lives at home. Ask by voice, in your own language. Nothing leaves the house.**
+
+*Saral* (सरल) means "simple." Saral is built for people who will never open a terminal: parents and grandparents, newcomers to a country, and small-business owners. It helps with the confusing moments that come up every week:
+
+- **Is this a scam?** Forward or paste a suspicious SMS, WhatsApp message or email. Saral lists the warning signs in plain words and always tells you how to check safely.
+- **What does this paper say?** Take a photo of a bill, notice or letter. Saral finds the amounts, dates and phone numbers exactly as written, and explains what it asks you to do.
+
+Coming next: *help me reply*, *talk for me* (live translation) and *help me fill this form*. See the [roadmap](https://github.com/Fullcountr/saral/blob/main/docs/ROADMAP.md).
+
+> **Status: v0.1 alpha.** The core loop works end to end. It is not yet ready for unsupervised use by vulnerable people. See [what v0.1 must prove](#what-v01-must-prove).
+
+---
+
+## The privacy promise
+
+1. **Everything runs on your own computer.** The AI model, speech recognition and photo reading all run locally. The only internet use is downloading the AI model once, from Hugging Face, which sends none of your data.
+2. **Saral refuses to send your data off the machine.** The model connection only accepts this computer or your home network, unless you explicitly change that.
+3. **No content telemetry, ever.** Photos are processed in memory and not stored.
+4. **Free and open source** under Apache 2.0. Anyone can read the code and check these claims.
+
+## How it works
+
+```
+ Phone (web app, paired by QR)  ──home Wi-Fi──▶  Saral on a laptop or mini-PC
+                                                   │
+                                                   ├─ Read:   photo → text (Tesseract OCR)
+                                                   ├─ Facts:  amounts, dates, phones, links (rules, not AI)
+                                                   ├─ Think:  explanation in your language (Qwen3, bundled llama.cpp engine)
+                                                   ├─ Guard:  no invented numbers, every point quoted from the paper
+                                                   └─ Safety: fixed warnings, verification steps, emergency numbers
+```
+
+**The design rule: extract facts deterministically, let the AI explain them, and let the human decide.**
+
+- Numbers and dates are pulled out by rules, never by the model, and shown exactly as written.
+- Every explanation point must quote the paper. Points that can't be matched are dropped.
+- If the model writes a number that isn't on the paper, the answer is rejected and retried; if it still fails, Saral shows only the exact facts.
+- Scam warnings and "how to check safely" steps are written by people, in every language, and always shown. Saral never says a message is "safe."
+- Emergency phrases ("chest pain", "can't breathe", self-harm) get a fixed response with the right phone numbers before any AI runs.
+
+## Install (for families)
+
+No terminal, no extra programs. Saral includes its own AI engine.
+
+1. **Download** the installer for your computer from the [Releases page](https://github.com/Fullcountr/saral/releases): `Saral-Setup.exe` for Windows, the `.dmg` for a Mac.
+2. **Double-click it.** A setup screen opens in your browser, in your language. It checks the computer and downloads Saral's AI once (1.3–5 GB, resumes if the internet drops). No internet? Use a Saral USB stick instead.
+3. **Scan the code** on the screen with your phone's camera, then tap "Add to Home screen."
+
+That's it. Saral starts by itself whenever the computer turns on.
+
+> The first public installers will appear with the first release. Until they are code-signed, Windows may say "Windows protected your PC" (click **More info → Run anyway**), and on a Mac you right-click Saral and choose **Open** the first time. Windows may also ask to allow Saral on your network: choose **Allow**, so your phone can reach it.
+
+**Best for elders:** a family member sets it up once, or uses a preloaded Saral box (planned), so the person using Saral never installs anything.
+
+## Install with pip (for tech-comfortable users)
+
+```bash
+pip install saral
+saral            # opens the setup screen; it downloads the AI engine and model for you
+```
+
+To read photos of papers, also install Tesseract: `brew install tesseract tesseract-lang` (Mac), `sudo apt install tesseract-ocr tesseract-ocr-hin tesseract-ocr-tel` (Ubuntu), or the [Windows installer](https://github.com/UB-Mannheim/tesseract/wiki). The setup screen reminds you if it's missing.
+
+## For developers
+
+```bash
+git clone https://github.com/Fullcountr/saral.git
+cd saral
+pip install -e ".[dev,ocr]"
+
+saral            # the same one-click app: opens the setup screen
+saral doctor     # what this computer has and which model tier it gets
+```
+
+For AI answers during development, put llama.cpp's `llama-server` on your PATH (or set `SARAL_LLAMA_SERVER`). You can also use an existing Ollama install with `saral serve --backend ollama`, or run with no model at all: `saral serve --backend rules`.
+
+Build the installer locally with `pip install -e ".[ocr,build]"`, then `python packaging/fetch_llama.py --platform <win-x64|macos-arm64|macos-x64|linux-x64>`, `python packaging/bundle_tesseract.py` and `python packaging/build.py`. Pushing a `v*` tag builds all installers in GitHub Actions ([release.yml](https://github.com/Fullcountr/saral/blob/main/.github/workflows/release.yml)).
+
+## Hardware tiers
+
+`saral doctor` measures the machine and picks a tier automatically.
+
+| Tier | Typical machine | Model | Good for |
+| --- | --- | --- | --- |
+| Lite | 8 GB laptop, no GPU | `qwen3:1.7b` | Scam checks, short documents; slower |
+| Standard | 16 GB laptop or mini-PC | `qwen3:4b` | Everything in v0.1 at usable speed |
+| Plus | 32 GB, Apple Silicon or 12 GB+ GPU | `qwen3:8b` | Long documents, better explanations |
+
+## Languages
+
+v0.1 ships English, Hindi (हिन्दी), Telugu (తెలుగు) and Spanish. Safety text in Hindi, Telugu and Spanish **needs review by native speakers**; please help (see [CONTRIBUTING.md](https://github.com/Fullcountr/saral/blob/main/CONTRIBUTING.md)).
+
+## Testing and evals
+
+```bash
+pytest                         # unit, API, safety and "no network" tests
+saral eval --backend rules     # deterministic evals (runs in CI)
+saral eval --backend ollama    # full evals with your local model
+```
+
+The eval sets in [`evals/`](https://github.com/Fullcountr/saral/blob/main/evals/) are the most valuable part of the project: real-world-style messages and documents per language with ground truth. Contributions of new (anonymized or synthetic) cases are very welcome.
+
+## What v0.1 must prove
+
+Before adding features, v0.1 has to pass its gate:
+
+- Amounts and dates extracted exactly in ≥99% of eval documents, with zero changed digits in explanations.
+- Zero outbound network connections during a full session (enforced by a test).
+- Five real testers finish "is this a scam?" and "explain this paper" without help.
+
+## Project docs
+
+- [Architecture](https://github.com/Fullcountr/saral/blob/main/docs/ARCHITECTURE.md)
+- [Edge cases and how Saral handles them](https://github.com/Fullcountr/saral/blob/main/docs/EDGE_CASES.md)
+- [Roadmap](https://github.com/Fullcountr/saral/blob/main/docs/ROADMAP.md)
+- [Third-party models and licenses](https://github.com/Fullcountr/saral/blob/main/docs/THIRD_PARTY.md)
+- [Security policy](https://github.com/Fullcountr/saral/blob/main/SECURITY.md) · [Trademark policy](https://github.com/Fullcountr/saral/blob/main/TRADEMARK.md) · [Contributing](https://github.com/Fullcountr/saral/blob/main/CONTRIBUTING.md)
+
+## License
+
+Code: [Apache License 2.0](https://github.com/Fullcountr/saral/blob/main/LICENSE). The name "Saral" and its logo are covered by the [trademark policy](https://github.com/Fullcountr/saral/blob/main/TRADEMARK.md).
+
+Saral explains and drafts; it does not give legal, medical or financial advice, and it never acts on your behalf.

saral-0.1.0/README.md

@@ -0,0 +1,125 @@
+# Saral
+
+**A private AI helper that lives at home. Ask by voice, in your own language. Nothing leaves the house.**
+
+*Saral* (सरल) means "simple." Saral is built for people who will never open a terminal: parents and grandparents, newcomers to a country, and small-business owners. It helps with the confusing moments that come up every week:
+
+- **Is this a scam?** Forward or paste a suspicious SMS, WhatsApp message or email. Saral lists the warning signs in plain words and always tells you how to check safely.
+- **What does this paper say?** Take a photo of a bill, notice or letter. Saral finds the amounts, dates and phone numbers exactly as written, and explains what it asks you to do.
+
+Coming next: *help me reply*, *talk for me* (live translation) and *help me fill this form*. See the [roadmap](https://github.com/Fullcountr/saral/blob/main/docs/ROADMAP.md).
+
+> **Status: v0.1 alpha.** The core loop works end to end. It is not yet ready for unsupervised use by vulnerable people. See [what v0.1 must prove](#what-v01-must-prove).
+
+---
+
+## The privacy promise
+
+1. **Everything runs on your own computer.** The AI model, speech recognition and photo reading all run locally. The only internet use is downloading the AI model once, from Hugging Face, which sends none of your data.
+2. **Saral refuses to send your data off the machine.** The model connection only accepts this computer or your home network, unless you explicitly change that.
+3. **No content telemetry, ever.** Photos are processed in memory and not stored.
+4. **Free and open source** under Apache 2.0. Anyone can read the code and check these claims.
+
+## How it works
+
+```
+ Phone (web app, paired by QR)  ──home Wi-Fi──▶  Saral on a laptop or mini-PC
+                                                   │
+                                                   ├─ Read:   photo → text (Tesseract OCR)
+                                                   ├─ Facts:  amounts, dates, phones, links (rules, not AI)
+                                                   ├─ Think:  explanation in your language (Qwen3, bundled llama.cpp engine)
+                                                   ├─ Guard:  no invented numbers, every point quoted from the paper
+                                                   └─ Safety: fixed warnings, verification steps, emergency numbers
+```
+
+**The design rule: extract facts deterministically, let the AI explain them, and let the human decide.**
+
+- Numbers and dates are pulled out by rules, never by the model, and shown exactly as written.
+- Every explanation point must quote the paper. Points that can't be matched are dropped.
+- If the model writes a number that isn't on the paper, the answer is rejected and retried; if it still fails, Saral shows only the exact facts.
+- Scam warnings and "how to check safely" steps are written by people, in every language, and always shown. Saral never says a message is "safe."
+- Emergency phrases ("chest pain", "can't breathe", self-harm) get a fixed response with the right phone numbers before any AI runs.
+
+## Install (for families)
+
+No terminal, no extra programs. Saral includes its own AI engine.
+
+1. **Download** the installer for your computer from the [Releases page](https://github.com/Fullcountr/saral/releases): `Saral-Setup.exe` for Windows, the `.dmg` for a Mac.
+2. **Double-click it.** A setup screen opens in your browser, in your language. It checks the computer and downloads Saral's AI once (1.3–5 GB, resumes if the internet drops). No internet? Use a Saral USB stick instead.
+3. **Scan the code** on the screen with your phone's camera, then tap "Add to Home screen."
+
+That's it. Saral starts by itself whenever the computer turns on.
+
+> The first public installers will appear with the first release. Until they are code-signed, Windows may say "Windows protected your PC" (click **More info → Run anyway**), and on a Mac you right-click Saral and choose **Open** the first time. Windows may also ask to allow Saral on your network: choose **Allow**, so your phone can reach it.
+
+**Best for elders:** a family member sets it up once, or uses a preloaded Saral box (planned), so the person using Saral never installs anything.
+
+## Install with pip (for tech-comfortable users)
+
+```bash
+pip install saral
+saral            # opens the setup screen; it downloads the AI engine and model for you
+```
+
+To read photos of papers, also install Tesseract: `brew install tesseract tesseract-lang` (Mac), `sudo apt install tesseract-ocr tesseract-ocr-hin tesseract-ocr-tel` (Ubuntu), or the [Windows installer](https://github.com/UB-Mannheim/tesseract/wiki). The setup screen reminds you if it's missing.
+
+## For developers
+
+```bash
+git clone https://github.com/Fullcountr/saral.git
+cd saral
+pip install -e ".[dev,ocr]"
+
+saral            # the same one-click app: opens the setup screen
+saral doctor     # what this computer has and which model tier it gets
+```
+
+For AI answers during development, put llama.cpp's `llama-server` on your PATH (or set `SARAL_LLAMA_SERVER`). You can also use an existing Ollama install with `saral serve --backend ollama`, or run with no model at all: `saral serve --backend rules`.
+
+Build the installer locally with `pip install -e ".[ocr,build]"`, then `python packaging/fetch_llama.py --platform <win-x64|macos-arm64|macos-x64|linux-x64>`, `python packaging/bundle_tesseract.py` and `python packaging/build.py`. Pushing a `v*` tag builds all installers in GitHub Actions ([release.yml](https://github.com/Fullcountr/saral/blob/main/.github/workflows/release.yml)).
+
+## Hardware tiers
+
+`saral doctor` measures the machine and picks a tier automatically.
+
+| Tier | Typical machine | Model | Good for |
+| --- | --- | --- | --- |
+| Lite | 8 GB laptop, no GPU | `qwen3:1.7b` | Scam checks, short documents; slower |
+| Standard | 16 GB laptop or mini-PC | `qwen3:4b` | Everything in v0.1 at usable speed |
+| Plus | 32 GB, Apple Silicon or 12 GB+ GPU | `qwen3:8b` | Long documents, better explanations |
+
+## Languages
+
+v0.1 ships English, Hindi (हिन्दी), Telugu (తెలుగు) and Spanish. Safety text in Hindi, Telugu and Spanish **needs review by native speakers**; please help (see [CONTRIBUTING.md](https://github.com/Fullcountr/saral/blob/main/CONTRIBUTING.md)).
+
+## Testing and evals
+
+```bash
+pytest                         # unit, API, safety and "no network" tests
+saral eval --backend rules     # deterministic evals (runs in CI)
+saral eval --backend ollama    # full evals with your local model
+```
+
+The eval sets in [`evals/`](https://github.com/Fullcountr/saral/blob/main/evals/) are the most valuable part of the project: real-world-style messages and documents per language with ground truth. Contributions of new (anonymized or synthetic) cases are very welcome.
+
+## What v0.1 must prove
+
+Before adding features, v0.1 has to pass its gate:
+
+- Amounts and dates extracted exactly in ≥99% of eval documents, with zero changed digits in explanations.
+- Zero outbound network connections during a full session (enforced by a test).
+- Five real testers finish "is this a scam?" and "explain this paper" without help.
+
+## Project docs
+
+- [Architecture](https://github.com/Fullcountr/saral/blob/main/docs/ARCHITECTURE.md)
+- [Edge cases and how Saral handles them](https://github.com/Fullcountr/saral/blob/main/docs/EDGE_CASES.md)
+- [Roadmap](https://github.com/Fullcountr/saral/blob/main/docs/ROADMAP.md)
+- [Third-party models and licenses](https://github.com/Fullcountr/saral/blob/main/docs/THIRD_PARTY.md)
+- [Security policy](https://github.com/Fullcountr/saral/blob/main/SECURITY.md) · [Trademark policy](https://github.com/Fullcountr/saral/blob/main/TRADEMARK.md) · [Contributing](https://github.com/Fullcountr/saral/blob/main/CONTRIBUTING.md)
+
+## License
+
+Code: [Apache License 2.0](https://github.com/Fullcountr/saral/blob/main/LICENSE). The name "Saral" and its logo are covered by the [trademark policy](https://github.com/Fullcountr/saral/blob/main/TRADEMARK.md).
+
+Saral explains and drafts; it does not give legal, medical or financial advice, and it never acts on your behalf.

saral-0.1.0/docs/ARCHITECTURE.md

@@ -0,0 +1,62 @@
+# Architecture
+
+Saral is one Python process on a laptop or mini-PC at home, plus a web app the phone opens over home Wi-Fi.
+
+```
+Phone web app (src/saral/web)            Home box (src/saral)
+┌────────────────────────────┐          ┌───────────────────────────────────────────────┐
+│ big buttons, 4 languages   │  HTTP +  │ server.py   FastAPI, pairing token, one model │
+│ camera → downscaled JPEG   │  token   │             request at a time (queue)          │
+│ push-to-talk (needs HTTPS) │ ───────▶ │ safety.py   emergency phrases → fixed answer   │
+│ read-aloud: on-device      │          │ ocr.py      photo quality check + Tesseract    │
+│ voices only                │          │ extract.py  amounts, dates, phones, links      │
+└────────────────────────────┘          │ jobs/       scam.py, explain.py                │
+                                        │ guard.py    no new numbers, quotes must match  │
+                                        │ engine.py   bundled llama-server (127.0.0.1)   │
+                                        │ models.py   one-time verified model download   │
+                                        │ setup_api   first-run screen, QR, auto-start   │
+                                        │ i18n.py     human-written safety text          │
+                                        └───────────────────────────────────────────────┘
+```
+
+## Request flow: "What does this paper say?"
+
+1. Phone downscales the photo and posts it with the user's language.
+2. `safety.emergency_response` checks the user's own words (voice or question) first.
+3. `ocr.quality_issues` checks for dark, glary, blurry or tiny photos; `ocr.ocr_image` reads it.
+4. `extract.extract` pulls amounts, dates, phones, links and emails with their exact text and line.
+5. `jobs.scam.check_scam` runs on the document too, so a scam letter gets a warning.
+6. The model is asked for JSON: summary, points and actions, each with an exact quote.
+7. `guard.new_numbers` rejects any answer with numbers not on the paper; `guard.quote_in_source` drops points whose quotes don't match. One retry with the reason; then fall back to facts only.
+8. The phone renders with `textContent` only (no HTML from the model) and shows exact facts beside the explanation.
+
+## Request flow: "Is this a scam?"
+
+Rules in `jobs/scam.py` score twelve kinds of warning sign in four languages, plus link checks (shorteners, raw IPs, risky endings, brand look-alikes). The model may add known signs and a short explanation, but it can never lower the level, its explanation is dropped if it contains numbers or the word "safe," and the human-written verification step is always shown. Levels are `high`, `some` and `none_found`; there is no "safe."
+
+## Why these choices
+
+| Choice | Reason |
+| --- | --- |
+| Rules for numbers, model for words | A wrong amount or deadline is the worst failure; models garble numbers across languages. |
+| Fixed safety text in `i18n.py` | Warnings must be correct every time, in every language. |
+| Bundled llama.cpp + Qwen3 | Nothing extra to install; permissive licenses; runs on CPU; 119 training languages. |
+| Tesseract first | Runs on any CPU; PaddleOCR/Docling are the upgrade path for forms and tables. |
+| Plain JS, no build, no CDN | The app must load with no internet and be auditable. |
+| Pairing token | Shared housing and hostels have shared Wi-Fi. |
+
+## Model backends
+
+`llm.Backend` has four implementations: `LlamaCppBackend` (default: the bundled engine in `engine.py`), `OllamaBackend` (developers with Ollama), `RulesBackend` (no model, deterministic output only) and `FakeBackend` (tests). `check_local_url` refuses model hosts outside loopback or private network ranges unless `allow_remote_model_host` is set.
+
+## One-click app and first-run setup
+
+`app.py` is what the desktop icon runs. It keeps a fixed port so paired phones keep working, opens the browser on `/setup` the first time and on `/` afterwards, and if Saral is already running just opens the window again. The setup screen (`web/setup.html`) walks through: language → computer check → model download (or USB offline pack) → engine start → phone QR code and "start with this computer" (on by default). Setup endpoints answer only requests from the computer itself, never from phones.
+
+The model download (`models.py`) is the only internet use. It only fetches from Hugging Face, resumes partial downloads with HTTP Range, and checks the GGUF header, the size and the SHA-256 that Hugging Face publishes; a mismatch deletes the file.
+
+## Reusing what is already on the computer
+
+`discover.py` looks for models other apps downloaded: Ollama (through its manifests to the model blob), LM Studio, Jan, the Hugging Face cache, and Downloads/Desktop. It reads each file's GGUF header and accepts only Qwen3 files of the sizes Saral is tested with (1.7B, 4B, 8B), so a different model can't silently lower answer quality. A chosen file is recorded in `model_links.json` and run where it is: no copy, and Saral never deletes it, because it belongs to the other app. If the file disappears later, setup simply offers a download again.
+
+A running Ollama with a matching model can be used instead of Saral's engine; the choice is saved in `engine_choice.json`. For the engine itself, `binaries.llama_server_info()` prefers the bundled copy, then Saral's own download, then a system `llama-server` only if its build number is recent enough. A system engine that fails to start is remembered and skipped, so setup fetches Saral's own.