Celeste AI

The primitive layer for multi-modal AI

All modalities. All providers. One interface.

Primitives, not frameworks.

Quick Start • Request Provider

🚀 This is the v1 Beta release. We're validating the new architecture before the stable v1.0 release. Feedback welcome!

Celeste AI

Type-safe, modality/provider-agnostic primitives.

Unified Interface: One API for OpenAI, Anthropic, Gemini, Mistral, and 14+ others.
True Multi-Modal: Text, Image, Audio, Video, Embeddings, Search —all first-class citizens.
Type-Safe by Design: Full Pydantic validation and IDE autocomplete.
Zero Lock-In: Switch providers instantly by changing a single config string.
Primitives, Not Frameworks: No agents, no chains, no magic. Just clean I/O.
Lightweight Architecture: No vendor SDKs. Pure, fast HTTP.

🚀 Quick Start

import celeste

# One SDK. Every modality. Any provider.
text   = await celeste.text.generate("Explain quantum computing", model="claude-opus-4-5")
image  = await celeste.images.generate("A serene mountain lake at dawn", model="flux-2-pro")
speech = await celeste.audio.speak("Welcome to the future", model="eleven_v3")
video  = await celeste.videos.analyze(video_file, prompt="Summarize this clip", model="gemini-3-pro-preview")
embeddings = await celeste.text.embed(["lorep ipsum", "dolor sit amet"], model="gemini-embedding-001")

15+ providers. Zero lock-in.

and many more

Missing a provider? Request it – ⚡ we ship fast.

Operations by Domain

Action	Text	Images	Audio	Video
Generate	✓	✓	○	✓
Edit	—	✓	—	—
Analyze	—	✓	✓	✓
Upscale	—	○	—	○
Speak	—	—	✓	—
Transcribe	—	—	✓	—
Embed	✓	○	—	○

_{✓ Available · ○ Planned}

🔄 Switch providers in one line

from pydantic import BaseModel

class User(BaseModel):
    name: str
    age: int

# Model IDs
anthropic_model_id = "claude-4-5-sonnet"
google_model_id = "gemini-2.5-flash"

# ❌ Anthropic Way
from anthropic import Anthropic
import json

client = Anthropic()
response = client.messages.create(
    model=anthropic_model_id,
    messages=[
        {"role": "user",
         "content": "Extract user info: John is 30"}
    ],
    output_format={
        "type": "json_schema",
        "schema": User.model_json_schema()
    }
)
user_data = json.loads(response.content[0].text)

# ❌ Google Gemini Way
from google import genai
from google.genai import types

client = genai.Client()
response = await client.aio.models.generate_content(
    model=gemini_model_id,
    contents="Extract user info: John is 30",
    config=types.GenerateContentConfig(
        response_mime_type="application/json",
        response_schema=User
    )
)
user = response.parsed

# ✅ Celeste Way
import celeste

response = await celeste.text.generate(
    "Extract user info: John is 30",
    model=google_model_id,  # <--- Choose any model from any provider
    output_schema=User,  # <--- Unified parameter working across all providers
)
user = response.content  # Already parsed as User instance

⚙️ Advanced: Create Client

For explicit configuration or client reuse, use create_client with modality + operation. This is modality-first: you choose the output type and operation explicitly.

from celeste import create_client, Modality, Operation, Provider

client = create_client(
    modality=Modality.TEXT,
    operation=Operation.GENERATE,
    provider=Provider.OLLAMA,
    model="llama3.2",
)
response = await client.generate("Extract user info: John is 30", output_schema=User)

capability is still supported but deprecated. Prefer modality + operation.

🪶 Install

uv add celeste-ai
# or
pip install celeste-ai

🔧 Type-Safe by Design

# Full IDE autocomplete
import celeste

response = await celeste.text.generate(
    "Explain AI",
    model="gpt-4o-mini",
    temperature=0.7,    # ✅ Validated (0.0-2.0)
    max_tokens=100,     # ✅ Validated (int)
)

# Typed response
print(response.content)              # str (IDE knows the type)
print(response.usage.input_tokens)   # int
print(response.metadata["model"])     # str

Catch errors before production.

🤝 Contributing

We welcome contributions! See CONTRIBUTING.md.

Request a provider: GitHub Issues Report bugs: GitHub Issues

📄 License

MIT license – see LICENSE for details.

Get Started • Documentation • GitHub

Made with ❤️ by developers tired of framework lock-in

Release files for celeste-ai 0.11.0

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for celeste-ai 0.11.0
File	Size	Uploaded
celeste_ai-0.11.0.tar.gz	214.7 kB	Mar 27, 2026	Details

Built distribution (wheel)

Table of built distributions (wheels) for celeste-ai 0.11.0
File	Interpreter	ABI	Platform	Reset
celeste_ai-0.11.0-py3-none-any.whl 282.1 kB Mar 27, 2026	Python 3	none	any	Details

Total release size: 496.7 kB

Release files / celeste_ai-0.11.0.tar.gz

Download URL	celeste_ai-0.11.0.tar.gz
Size	214.7 kB
Tags	Source
SHA-256 checksum How to use checksums	`b2ec47a1ecd044f9a1baa5888c40ead95e31318db38117acb757100cd7d2ecf3`
BLAKE2b-256 checksum How to use checksums	`4b495b7404fd3f4c8c344b3da42a669e3f4ed69c4ea9bca2bb60b4709a95be02`
Upload date	Mar 27, 2026
Uploaded using Trusted Publishing? What is trusted publishing?	Yes
Uploaded via	`twine/6.1.0 CPython/3.13.7`

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

Identity

Publishing platform: GitHub Actions
Publishing repository: github.com/withceleste/celeste-python
Publishing commit: github.com/withceleste/celeste-python/tree/2522d8d3ec37926735d30f1aa8ac66ad407e6907

Workflow

Publishing configuration: github.com/withceleste/celeste-python/blob/2522d8d3ec37926735d30f1aa8ac66ad407e6907/.github/workflows/publish.yml
Publishing logs: github.com/withceleste/celeste-python/actions/runs/23642278652/attempts/7

Artifact

Subject: celeste_ai-0.11.0.tar.gz
SHA-256 checksum: b2ec47a1ecd044f9a1baa5888c40ead95e31318db38117acb757100cd7d2ecf3 Verifying attestations

Release files / celeste_ai-0.11.0-py3-none-any.whl

Download URL	celeste_ai-0.11.0-py3-none-any.whl
Size	282.1 kB
Tags	Python 3
SHA-256 checksum How to use checksums	`1f51b7ab162794076ab52e23b4b3268d4a772b75cf6db57ddce4af0f5d1705aa`
BLAKE2b-256 checksum How to use checksums	`d1ebec611f8be544a4a0b31714f1167453c56d9bb6762c0f7c18fa4ebf642e54`
Upload date	Mar 27, 2026
Uploaded using Trusted Publishing? What is trusted publishing?	Yes
Uploaded via	`twine/6.1.0 CPython/3.13.7`

Provenance

Identity

Publishing platform: GitHub Actions
Publishing repository: github.com/withceleste/celeste-python
Publishing commit: github.com/withceleste/celeste-python/tree/2522d8d3ec37926735d30f1aa8ac66ad407e6907

Workflow

Publishing configuration: github.com/withceleste/celeste-python/blob/2522d8d3ec37926735d30f1aa8ac66ad407e6907/.github/workflows/publish.yml
Publishing logs: github.com/withceleste/celeste-python/actions/runs/23642278652/attempts/7

Artifact

Subject: celeste_ai-0.11.0-py3-none-any.whl
SHA-256 checksum: 1f51b7ab162794076ab52e23b4b3268d4a772b75cf6db57ddce4af0f5d1705aa Verifying attestations