Skip to content

ASHA Documentation

v0.4.2 - mission-aware agent governance (ANCHOR) plus drop-in security and optimization for LLM applications.

ASHA masks PII, reduces tokens, and checks for injection patterns before prompts reach a model. ANCHOR wraps autonomous agents with mission contracts, action guards, and OS sandboxing. The root API is process, sanitize, optimize, Agent, and anchor; everything else lives in subpackages.

Developer preview - pin asha==0.4.2 in production. APIs may change before 1.0.0.

Start here

Guide Description
Getting Started Install and first example
Quickstart 5-minute walkthrough
ANCHOR Runtime Mission governance for agents
API Reference process, sanitize, optimize, Agent, anchor
Core Concepts Modes, results, policy
Examples Common patterns

How it works

Input prompt
    → Security (PII + threats)
    → Compile (internal IR → structured prompt)
    → Optimize (token compression)
    → ProcessResult

For LLM apps, add wrap_llm(client) or use Agent to preprocess before generation.

For autonomous agents, use anchor(agent) for mission-aware governance.

Public imports

# Root
from asha import process, sanitize, optimize, Agent, anchor

# Common advanced imports
from asha.integrations import wrap_llm
from asha.runtime.anchor import anchor_any
from asha.types import ProcessResult
from asha.core.policy_config import PolicyConfig
from asha.runtime import PromptProcessor

wrap_llm, Pipeline, Processor, and lazy root exports were removed in v0.4.1. See deprecations.md.

Documentation map

Essentials

Architecture

  • Architecture - package layout and layer rules
  • Pipeline - processing flow (3 engines, not 7 stages)
  • Prompt IR - internal representation (not public API)
  • Routing - Agent(routing_config=...)

Operations

Project

Install

pip install asha   # Python 3.10+
pip install -e ".[docs]" && mkdocs serve   # local docs