This is where Rylox gets really exciting.

I think **v0.1.0 should answer one question extremely well:**

> **"Given a repository and a task, produce the best possible context package."**

Everything after that is about making it **smarter, faster, broader, and more integrated**, not changing its core purpose.

---

# v1.1 — Multi-language support

Today:

```text id="4qfplg"
Python
```

Then add:

* JavaScript
* TypeScript
* Go
* Rust
* Java
* C++
* C#
* PHP
* Kotlin

via additional Tree-sitter grammars.

---

# v1.2 — More embedding providers

Support:

* Hugging Face
* Ollama
* OpenAI
* Voyage AI
* Jina AI
* Nomic
* BGE
* E5
* GTE

Users just change:

```toml id="9vqk5n"
provider = "ollama"
```

and Rylox adapts.

---

# v1.3 — Better relationship analysis

Instead of only:

```text id="d3k6cy"
calls
imports
```

understand:

* inheritance
* overrides
* interfaces
* implementations
* decorators
* dependency injection
* FastAPI routes
* Django models
* SQLAlchemy models

---

# v1.4 — MCP Server

Instead of

```bash id="o2wfq0"
rylox context
```

an AI assistant can ask Rylox directly:

```text id="s79g2w"
Give me context for checkout.
```

Rylox becomes an MCP server.

---

# v1.5 — IDE integration

VS Code

JetBrains

Neovim

Cursor

Continue

Instead of manually running CLI.

---

# v1.6 — Watch mode

```bash id="rzzhcf"
rylox watch
```

Automatically re-index changed files.

---

# v1.7 — Better ranking

Learn from:

* clicks
* accepted answers
* edits

Instead of static BM25 + RRF.

---

# v1.8 — Smarter context assembly

Instead of

```text id="z6bny5"
top 20 chunks
```

Rylox starts reasoning:

```text id="v8ydke"
Need constructor

Need interface

Need implementation

Need tests

Need config
```

before assembling.

---

# v2.0 — Learning-based retrieval

This is where custom models become interesting.

Instead of BM25 + embeddings only:

```text id="xwfk1q"
Task

↓

Small ranking model

↓

Best context package
```

Maybe a fine-tuned reranker.

---

# v2.x ideas

* Distributed indexing
* Monorepo support
* Git history awareness
* Bug localization
* Automatic issue-to-context generation
* PR review context generation
* Agent memory
* Semantic diffing
* Repository summarization
* Cloud index synchronization
* Team knowledge graph

---

## The important thing

Notice that **none** of these invalidate v0.1.0.

If someone installs Rylox v0.1.0 and later upgrades to v1.8, their workflow is still:

```bash id="bqstcz"
rylox index

rylox context "Why is checkout slow?"
```

The commands stay the same.

Only the quality of the results improves.

That's exactly what you want from a well-designed tool: a stable interface with increasingly capable internals.

So in my mind:

* **v0.1.0** = A complete, usable local repository context engine.
* **v1.x** = Broader language support, integrations, and smarter retrieval.
* **v2.x** = Intelligent ranking, adaptive retrieval, and enterprise-scale capabilities.

That progression feels natural and gives Rylox plenty of room to grow without forcing users to relearn how to use it.
