Skip to content

Processing Flow

ASHA v0.4.2 - three engines, one orchestrator.

The old 7-stage Pipeline class was removed. PromptProcessor runs three core engines in order.

Flow

User prompt
    │
    ▼
Security (run_security)
    │  PII detection, injection checks, masking
    ▼
Compile (compile_prompt)
    │  Internal IR → structured prompt (IR not public)
    ▼
Optimize (optimize_tokens)
    │  MSDPC token compression
    ▼
ProcessResult

mode="off" skips all engines via the policy gate.

Entry points

Function Engines run
process() security → compile → optimize
sanitize() security only
optimize() optimize only 
from asha import process, sanitize, optimize

process("prompt with john@x.com")
sanitize("prompt with john@x.com")
optimize("long verbose prompt please analyze")

Orchestrator

from asha.runtime import PromptProcessor

processor = PromptProcessor()
result = processor.run("prompt", mode="balanced")

Stage control uses ExecutionProfile or PolicyConfig - not security= / compile= / optimize= kwargs.

Agent adds LLM generation

Agent.run(prompt)
  → PromptProcessor (preprocess)
  → Adapter.generate(processed prompt)
  → str or AgentResult

Model selection for multi-task apps:

Agent(routing_config={"chat": "gpt-4o-mini", "code": "gpt-4o"})

See routing.md.

Tracing

result = process("prompt", trace=True, debug=True)
print(result.trace)
print(result.diff)

Uses core/trace_context.py - not the removed ASHADebugger.

Legacy dict shape

Opt-in only:

from asha.compat.legacy_results import to_legacy_pipeline_dict

legacy = to_legacy_pipeline_dict(
    process("prompt", include_legacy_detail=True)
)

routing_decision appears only in legacy dicts (placeholder for compat).