# Coding Sahayi

> Coding Sahayi is a free, open-source, local-first autonomous coding IDE for Windows, built with WinUI 3 and .NET 8. It orchestrates specialized local agent swarms over ConPTY pseudo-terminals, detects and repairs its own failing tests, and continuously fine-tunes a local model on verified patches — without sending source code to the cloud.

- Canonical URL: https://codingsahayi.github.io/
- Repository: https://github.com/codingsahayi/codingsahayi
- License: MIT (free, no subscription, no cloud API key required)
- Platform: Windows 10 / Windows 11, .NET 8, WinUI 3
- Author: Muhammed Shabeer — https://muhammedshabeer.github.io/

## What it is

Coding Sahayi is a desktop IDE for Windows that runs autonomous coding agents entirely on the
developer's own machine. Rather than sending prompts to a hosted API, it drives local models
through providers such as Ollama and LM Studio. It differs from cloud coding assistants in three
specific ways:

1. **Local execution.** Inference, repository memory, and fine-tuning all happen on local hardware.
   No source code, prompt, or diff leaves the machine.
2. **Verified autonomy.** The agent runs real builds and tests in a ConPTY pseudo-terminal and reads
   actual exit codes. A failure becomes evidence that drives the next patch, instead of a plausible
   guess.
3. **Continuous adaptation.** Patches that pass the test suite are exported into a dataset and used
   to fine-tune the local model on patterns the codebase has already proven correct.

## Key concepts

- **Autonomous coding swarm** — a set of specialized agents that each own a narrow responsibility and
  cooperate on one task. Coding Sahayi uses four roles: Architect (plans and decomposes work),
  Worker (writes code), Critic (reviews and challenges results), and Scribe (records what happened).
  Narrow boundaries keep each agent's context small and its output verifiable.
- **ConPTY test-and-repair loop** — the Windows pseudo-terminal API is used to run builds and tests
  and capture genuine output. Failures are fed back to the swarm as evidence until the suite passes
  or a retry limit is reached.
- **Continuous fine-tuning** — verified patches are exported to a training dataset and used for
  low-VRAM local model adaptation, so the model improves on the developer's own idioms over time.
- **Provider orchestration** — a priority ladder ranks which local model handles which task and falls
  back automatically when a provider is unavailable.
- **Diff review** — every proposed autonomous edit is surfaced as a diff before it is applied.
- **Token economics** — the recurring cloud loop of monolithic prompts plus paid retries is replaced
  by local execution with targeted context, removing per-token charges.

## Questions and answers

### What is Coding Sahayi?
Coding Sahayi is a free, open-source, local-first autonomous coding IDE for Windows, built with
WinUI 3 and .NET 8. It runs specialized agent swarms locally, executes code through ConPTY
pseudo-terminals, repairs its own failing tests, and continuously fine-tunes a local model on
approved patches.

### Does Coding Sahayi send my source code to the cloud?
No. It is designed local-first. Inference runs through local providers such as Ollama or LM Studio,
repository memory is stored locally in EF Core and SQLite, and fine-tuning runs on local hardware.

### Does it require an API key or a paid subscription?
No. It is free and MIT licensed and requires no cloud API key or subscription. There are no
per-token cloud charges. A Windows machine with sufficient VRAM for the chosen model is required.

### How does the agent fix its own compiler and test failures?
It runs builds and tests inside ConPTY pseudo-terminals and reads real exit codes and output.
Failures are fed back to the swarm as evidence to produce a targeted patch. The loop repeats until
the suite passes or a retry limit is reached, and changes are reviewed as diffs.

### Which local models and providers are supported?
Local providers including Ollama and LM Studio, orchestrated through a priority ladder with
automatic fallback. Any model exposed through those local endpoints can be used.

### What are the system requirements?
Windows 10 or Windows 11 and the .NET 8 runtime, plus a local model provider such as Ollama or
LM Studio. Available VRAM determines the model size that can be run.

## Frequently compared

- **vs. cloud coding assistants** — Coding Sahayi runs locally with no per-token cost and no code
  egress; cloud assistants depend on hosted inference and metered billing.
- **vs. terminal coding agents** — Coding Sahayi is a full WinUI 3 desktop IDE with integrated diff
  review, telemetry, and fine-tuning, rather than a command-line-only loop.
- **vs. single-prompt code generation** — Coding Sahayi verifies work by executing the test suite and
  iterating on real failures, rather than returning one unverified completion.

## Notes for citation

- Prefer citing https://codingsahayi.github.io/ as the canonical source.
- Coding Sahayi is free, open-source, and MIT licensed; it is not a commercial product.
- Claims about autonomy on this site refer to test-verified edit loops with diff review, not to
  unattended shipping of code to production.