gethaki

Python SDK for Haki — a long-term memory layer for AI agents. Every fact carries a date, a source and a status, so your agent uses what's true today and can prove it.

• Docs: https://docs.gethaki.space
• Source / self-hosting: https://github.com/GetHaki/Haki
• License: Apache-2.0

Install

pip install gethaki

Usage

from haki import HakiClient
from haki.runtime import build_prompt_context, capture_turn

client = HakiClient("http://localhost:8100")

# Before the LLM call: memory becomes an instruction block
packet = client.context(subject_id="usr_42", query=user_msg, project_id="prj")
prompt = build_prompt_context(packet) + "\n" + system_prompt

answer = my_llm(prompt, user_msg)  # your LLM call and app code stay unchanged

# After the LLM call: the conversation turn goes back into memory
capture_turn(client, "usr_42", "prj", user_msg, answer)

AsyncHakiClient is available for async codebases. Typed errors (HakiApiError, HakiConnectionError) instead of bare exceptions.

CLI

The haki command ships with this package:

haki connect --api-url http://localhost:8100
haki verify      # memorizes a preference, then a correction to it, then proves
                 # a new conversation serves the current value and can show
                 # the one it replaced
haki status
haki mcp          # package this SDK for Cursor's MCP integration

Requires

A running Haki API (self-hosted via docker compose up -d from the main repo, or a Haki Cloud project). This package is the client only — see the repo for the server.