Metadata-Version: 2.4
Name: sovereigneg
Version: 0.1.0
Summary: Python SDK for SovereignEG — Sovereign AI Platform
Author-email: SovereignEG <dev@sovereigneg.ai>
License: MIT
Project-URL: Homepage, https://github.com/WaelAbouceo/SovereignEG
Project-URL: Documentation, https://github.com/WaelAbouceo/SovereignEG
Keywords: ai,llm,sovereign,egypt,inference
Classifier: Development Status :: 3 - Alpha
Classifier: Intended Audience :: Developers
Classifier: Programming Language :: Python :: 3
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
Requires-Python: >=3.9
Description-Content-Type: text/markdown
Requires-Dist: httpx>=0.25.0
Provides-Extra: dev
Requires-Dist: pytest; extra == "dev"
Requires-Dist: pytest-asyncio; extra == "dev"

# SovereignEG Python SDK

Official Python client for the SovereignEG sovereign AI platform.

## Install

```bash
pip install sovereigneg
```

## Quick Start

```python
from sovereigneg import SovereignEG

client = SovereignEG(
    api_key="sk-seg-your-key-here",
    base_url="http://localhost:8000",  # or https://api.sovereigneg.ai
)

# Chat
response = client.chat("What is sovereign AI?")
print(response.content)

# Streaming
for token in client.chat("Tell me about Egypt", stream=True):
    print(token, end="", flush=True)

# Embeddings
emb = client.embed("Hello world")
print(f"Dimension: {len(emb.embedding)}")
```

## Examples

See the [examples/](examples/) folder:

- `chat.py` — basic chat completion
- `streaming.py` — real-time token streaming
- `embeddings.py` — generate embeddings
- `rag.py` — upload documents and search

## Rate Limiting

All API responses include rate limit headers:

```
X-RateLimit-Limit: 60
X-RateLimit-Remaining: 42
X-RateLimit-Reset: 1771234567
```

When rate limited (HTTP 429), the response includes a `Retry-After` header and structured error:

```json
{
  "error": {
    "type": "rate_limit_exceeded",
    "message": "Rate limit exceeded. Your developer plan allows 60 requests per minute.",
    "plan": "developer",
    "limit": 60,
    "retry_after": 3.2
  }
}
```

**Handling rate limits in your code:**

```python
import time
from sovereigneg import SovereignEG

client = SovereignEG(api_key="sk-seg-...")

try:
    response = client.chat("Hello!")
except Exception as e:
    if hasattr(e, 'status_code') and e.status_code == 429:
        retry_after = e.response.headers.get("Retry-After", "1")
        time.sleep(float(retry_after))
        response = client.chat("Hello!")  # Retry
```

### Per-Plan Limits

| Plan | Chat RPM | Embeddings RPM | Burst | Concurrent |
|---|---|---|---|---|
| Free (Student) | 10 | 20 | 3 | 1 |
| Developer | 60 | 120 | 10 | 3 |
| SME (Business) | 200 | 500 | 30 | 10 |

## Automatic Retries

The SDK automatically retries on transient errors (429 rate limits, 5xx server errors, network timeouts) with exponential backoff:

```python
# Default: 2 retries. Customize:
client = SovereignEG(api_key="sk-seg-...", max_retries=5)
```

The SDK respects `Retry-After` headers from the server, so rate-limited requests wait the appropriate duration before retrying.

## API Reference

### `SovereignEG(api_key, base_url, timeout, tenant_id, max_retries)`

Create a client instance. `max_retries` defaults to `2`.

### `client.chat(message, model, system, temperature, max_tokens, stream)`

Send a chat completion. Returns `ChatCompletion` or `ChatStream`.

### `client.embed(input, model)`

Generate embeddings. Returns `Embedding` with `.embedding` or `.embeddings`.

### `client.upload(file_path, tenant_id)`

Upload a document for RAG. Returns `DocumentUpload`.

**Document ACL** — Control who can access uploaded documents:

```python
# Default: all tenant members can read
client.upload("report.pdf", tenant_id="your-tenant-id")

# Group-restricted: only engineering and finance groups
client.upload("report.pdf", tenant_id="your-tenant-id",
              acl_level="group", acl_groups="engineering,finance")

# Private: only the uploader can read
client.upload("report.pdf", tenant_id="your-tenant-id", acl_level="private")
```

ACL levels:
- `tenant` (default) — all members of the tenant can read
- `group` — only members of the specified groups can read
- `private` — only the document owner (uploader) can read

ACL is enforced on search, list, and delete. Admins can always access all documents.

### `client.search(query, top_k, tenant_id)`

Search documents. Returns list of `SearchResult`. Results are ACL-filtered — only documents the caller has permission to see are returned.
