Skip to content

CHANGELOG.md

Latest commit

 

History

History
231 lines (151 loc) · 7.63 KB

CHANGELOG.md

File metadata and controls

231 lines (151 loc) · 7.63 KB

Changelog

All notable changes to Cost Katana Python SDK will be documented in this file.

[Unreleased]

[2.5.4] - 2026-04-22

Added

  • CostKatanaClient.send_message: thinking, thinking_effort, and thinking_budget_tokens — enables Claude extended thinking on supported models; payload includes thinking with enabled, optional effort, and optional budgetTokens.
  • ck.ai(): same options via **options; SimpleResponse.thinking is set when the API returns reasoning content.

Changed

  • AWS Bedrock: alias llama-3.2-1b / aws_bedrock.llama_3_2_1b_instruct now maps to meta.llama4-scout-17b-instruct-v1:0 (replaces retired meta.llama3-2-1b-instruct-v1:0).

[2.5.3] - 2026-03-29

Documentation

  • README and QUICKSTART: PyPI install name cost-katana, table of contents, experimentation section, aligned examples.

[2.5.2] - 2026-03-27

Added

  • cost_katana.gateway: gateway_request_headers() for opt-out CostKatana-LLM-Security-Enabled / CostKatana-Output-Moderation-Enabled on direct gateway HTTP calls; GATEWAY_API_PREFIX.
  • CostKatanaClient.get_gateway_security_summary() — GET /api/gateway/security/summary (firewall + moderation aggregates).
  • README: gateway security defaults and security summary API.

Documentation

  • API paths: Confirmed the SDK uses https://api.costkatana.com with REST paths under /api/... (e.g. chat, templates) and OpenAI-compatible gateway calls under /api/gateway/v1/....

[2.5.1] - 2025-03-25

Added

  • README: AI Gateway (HTTP) section — OpenAI-compatible (/v1/chat/completions) and Anthropic (/v1/messages) examples with httpx, optional COSTKATANA_GATEWAY_URL and x-project-id, plus curl; link to costkatana-examples/2-gateway.

[2.5.0] - 2025-03-25

Added

  • auto_configure() — lazy global client initialization from COST_KATANA_API_KEY (used before ai(), chat(), track()).
  • from_env() — returns a CostKatanaClient built from environment (same two-variable contract as Config.from_env()).
  • track(entry) — manual cost logging without wiring AILogger; works with lazy env-based setup.
  • groq model namespace and Groq provider label in get_provider_from_model.
  • project_id on Config with env fallbacks: PROJECT_ID, COST_KATANA_PROJECT, COSTKATANA_PROJECT_ID.

Changed

  • Public env contract: only COST_KATANA_API_KEY (required) and PROJECT_ID (optional). Removed reading API_KEY, COST_KATANA_BASE_URL, COST_KATANA_DEFAULT_MODEL, and COST_KATANA_TIMEOUT from the environment — base URL (https://api.costkatana.com), default model, and timeouts are package constants.
  • get_global_client() calls auto_configure() so ai() / chat() work without an explicit configure() when the API key is in the environment.
  • CostKatanaClient sends optional x-project-id when project_id is set; warns once when PROJECT_ID is missing.
  • ai_logger singleton is lazy — initializes on first use with COST_KATANA_API_KEY / PROJECT_ID from the environment.
  • Duplicate mistral class in models_constants.py merged into a single definition.

[2.2.1] - 2025-01-19

📚 Documentation Updates

  • Clarified API Key Requirements: Updated documentation to clearly state that OpenAI and Gemini providers require user-provided API keys
  • Added prominent warnings that Cost Katana does not provide OpenAI or Google API keys
  • Updated environment variable documentation with detailed explanations for all providers
  • Added clear consequences for missing API keys
  • Improved consistency across all documentation files

🔧 Configuration

  • Enhanced environment variable documentation with user-provided key requirements:
    • OPENAI_API_KEY - Required for OpenAI models (GPT-4, GPT-3.5, etc.) - USER PROVIDED
    • GEMINI_API_KEY - Required for Google Gemini models (Gemini 2.5 Pro, Flash, etc.) - USER PROVIDED
    • AWS_ACCESS_KEY_ID / AWS_SECRET_ACCESS_KEY - Required for AWS Bedrock models (Claude, Nova, etc.) - USER PROVIDED
    • ANTHROPIC_API_KEY - Optional for direct Anthropic API access - USER PROVIDED
  • Added specific notes about which providers require user API keys
  • Updated README.md with comprehensive API key setup instructions
  • All API keys must be provided by users - Cost Katana does not include any provider API keys

[2.0.7] - 2025-01-23

📚 Documentation Updates

  • Updated README with refined examples repository links
  • Improved documentation clarity and structure
  • Enhanced package metadata

[2.0.6] - 2025-01-23

📚 Documentation Updates

  • Updated package metadata and documentation
  • Improved examples and usage guides
  • Enhanced README with better examples repository links

[2.0.5] - 2025-01-23

📚 Documentation Updates

  • Added prominent "More Examples" section in README linking to complete examples repository
  • Added direct links to costkatana-examples repository
  • Highlighted 300+ production-ready code samples across HTTP, TypeScript, Python, and frameworks
  • Special emphasis on Python SDK examples (Section 8) and FastAPI integration

🔗 Resources

[2.0.0] - 2025-01-XX

🚀 Major Release: Complete Simplification

Complete redesign to make using AI in Python as simple as possible.

✨ New Features

Ultra-Simple API

New ai() function - The easiest way to use AI:

import cost_katana as ck
response = ck.ai('gpt-4', 'Hello')
print(response.text)
print(f"Cost: ${response.cost}")

New chat() function - Simple chat sessions:

import cost_katana as ck
session = ck.chat('gpt-4')
session.send('Hello')
session.send('How are you?')
print(f"Total: ${session.total_cost}")

Auto-Configuration

  • Automatically detects API keys from environment
  • Works with COST_KATANA_API_KEY or provider keys directly
  • Zero setup if environment is configured
  • Smart error messages with actionable steps

SimpleResponse Object

  • Clean, simple response object
  • Direct access to text, cost, tokens
  • Bonus fields: cached, optimized, provider

💥 Breaking Changes

None! The traditional API (GenerativeModel) still works for backward compatibility.

You can still use:

model = ck.GenerativeModel('gpt-4')
response = model.generate_content('Hello')

Or use the new simple API:

response = ck.ai('gpt-4', 'Hello')

📚 Documentation

  • Completely rewritten README: Focus on simplicity
  • Updated QUICKSTART: 3-step process
  • New examples: Real-world use cases with simple API
  • Migration guides: From OpenAI, Anthropic, Google SDKs

🎯 Comparison

Before (v1.x)

import cost_katana as ck

ck.configure(api_key='dak_...')
model = ck.GenerativeModel('gpt-4')
response = model.generate_content('Hello')
print(response.text)

After (v2.0)

import cost_katana as ck

response = ck.ai('gpt-4', 'Hello')
print(response.text)

📦 Examples

New streamlined examples:

  • simple_examples.py - All basic patterns
  • basic_usage.py - Getting started
  • Updated all existing examples to show both APIs

🐛 Bug Fixes

  • Improved error messages
  • Better environment variable detection
  • More robust provider inference

[1.0.0] - 2024-XX-XX

Initial Release

  • Multi-provider support
  • Cost tracking
  • Unified API
  • Configuration management