Skip to content

Installation

For Users

uv add ace-framework

uv add 'ace-framework[all]'            # All optional features
uv add 'ace-framework[instructor]'     # Structured outputs (Instructor)
uv add 'ace-framework[langchain]'      # LangChain integration
uv add 'ace-framework[browser-use]'    # Browser automation
uv add 'ace-framework[claude-code]'    # Claude Code CLI integration
uv add 'ace-framework[claude-sdk]'     # Anthropic SDK integration steps
uv add 'ace-framework[observability]'  # Opik monitoring + cost tracking
uv add 'ace-framework[deduplication]'  # Skill deduplication (embeddings)
uv add 'ace-framework[transformers]'   # Local model support

For Contributors

git clone https://github.com/kayba-ai/agentic-context-engine
cd agentic-context-engine
uv sync  # Installs everything (10-100x faster than pip)

git clone https://github.com/kayba-ai/agentic-context-engine
cd agentic-context-engine
uv add -e .

Requirements

  • Python 3.12
  • An API key for your LLM provider

Configure Your LLM

The recommended way to set up your API keys and model selection:

ace setup

This interactive wizard validates your API key and model, then saves config to ace.toml (model names, safe to commit) and .env (API keys, gitignored). See Setup for full details.

Manual alternative

If you prefer not to use the wizard, set environment variables directly:

export OPENAI_API_KEY="sk-..."

Or create a .env file (add to .gitignore):

OPENAI_API_KEY=sk-...

Verify Installation

from ace import ACELiteLLM

# Uses ace.toml + .env from `ace setup`
agent = ACELiteLLM.from_setup()
print(agent.ask("Hello!"))

Or without ace setup:

agent = ACELiteLLM.from_model("gpt-4o-mini")
print(agent.ask("Hello!"))

Set Up Coding Agent Skills (Optional)

If you want the hosted kayba CLI or kayba setup, install the cloud extra first:

uv add 'ace-framework[cloud]'

Then, if you use Claude Code, install the Kayba pipeline skill:

kayba setup

This installs the evaluation pipeline skill to .claude/skills/ and prints CLI instructions. See Hosted API for details.