Hexify SDK

Config-driven Python SDK for building AI agents and RAG pipelines.

Import name: hexify · Pip name: hexify-ai · Python: 3.10+

v1 is a library, not a hosted API: config in, working agent out, in one Python call.

from hexify import Agent

agent = Agent(
    name="assistant",
    llm="claude-sonnet-5",
    system_prompt="You are a helpful assistant.",
)
print(agent.run("Hello!"))

Install

pip install hexify-ai

From this repo: pip install -e ".[dev]".

Quickstart (under 5 minutes)

1. Python 3.10+ and an Anthropic API key.
2. Set ANTHROPIC_API_KEY (export ANTHROPIC_API_KEY=sk-ant-... or, in PowerShell, $env:ANTHROPIC_API_KEY="sk-ant-...").
3. pip install hexify-ai
4. Run the snippet above, or:

python examples/basic_agent.py

Subsequent agent.run(...) calls on the same instance keep in-process conversation memory.

RAG + tools

from hexify import Agent, RAG

rag = RAG(
    source="./docs",
    embedder="openai",
    vectorstore="pgvector",
    pg_connection_string="postgresql://user:pass@localhost/hexify",
)
agent = Agent(
    name="support-bot",
    llm="claude-sonnet-5",
    rag=rag,
    tools=["web_search"],
    system_prompt="You are a support agent for Acme App.",
)
print(agent.run("How do I reset my password?"))

v1 models: claude-sonnet-5 (default), claude-haiku-4-5.
v1 vectorstore: pgvector only.
v1 built-in tools: web_search and send_email (deterministic stubs — no live web or SMTP).

RAG also needs OPENAI_API_KEY (embeddings) and a Postgres database with the vector extension.

CLI

hexify init
hexify run hexify.json --message "Hello"
hexify run hexify.json          # interactive prompt (quit / Ctrl-D to exit)

hexify.json is an AgentConfig file. Folder rag_source values are resolved relative to that file. Set HEXIFY_PG_CONNECTION_STRING when RAG is enabled.

Environment

Custom tools

from hexify.tools import tool

@tool(
    name="order_status",
    description="Look up an order by id.",
    input_schema={
        "type": "object",
        "properties": {"order_id": {"type": "string"}},
        "required": ["order_id"],
    },
)
def order_status(args: dict) -> str:
    return f"Order {args['order_id']} shipped."

Then pass tools=["order_status"] (and any built-ins) into Agent.

Logging

Each Agent.run emits one JSON line on the hexify.run logger: tokens, latency (ms), tool names, RAG hit count, and ok. hexify run prints these on stderr.

Examples

• examples/basic_agent.py — smallest working agent
• examples/rag_support_bot.py — RAG + web_search
• examples/hexify.json — CLI config

Docs

• User guide — config schema, CLI, tools, memory, limits
• v1 build spec — scope and public API contract

Develop

For the full contributor guide (branching, testing, releasing), see DEVELOPING.md. The short version:

git clone https://github.com/asad926/hexify-sdk
cd hexify-sdk
python -m venv .venv
. .venv/Scripts/Activate.ps1   # Windows PowerShell
pip install -e ".[dev]"
pre-commit install             # optional but recommended
ruff check .
pytest --cov=hexify

cp .env.example .env to start a local secrets file (git-ignored).

License

MIT — see LICENSE.