FAQ¶
ASHA v0.4.2
General¶
What is ASHA?¶
A drop-in layer that masks PII, checks prompt injection patterns, and compresses tokens before prompts reach LLM providers.
Is it production-ready?¶
For pinned pilots: yes, with monitoring. Architecture is complete in 0.4.2. For stable semver: not until 1.0.0. Use asha==0.4.2 and read developer-preview.md.
Python version?¶
3.10+ (3.10, 3.11, 3.12 in CI).
API¶
What can I import from the root?¶
from asha import process, sanitize, optimize, Agent
Nothing else - no wrap_llm, Pipeline, or Processor at root.
Where is wrap_llm?¶
from asha.integrations import wrap_llm
What does process() return?¶
A ProcessResult dataclass. str(result) returns result.output.
Does process() raise?¶
mode="balanced"(default): fail-open - degraded fallback,result.degraded=Truemode="strict": raisesASHAProcessingErroron total failure
How do I get metrics?¶
result = process("prompt")
print(result.metrics.token_reduction_pct)
print(result.security.pii_detected)
No return_metrics=True - use typed fields.
How do I set pii_mode or reversible?¶
from asha.core.policy_config import PolicyConfig
process(prompt, policy=PolicyConfig(pii_mode="hybrid", reversible=True))
Top-level pii_mode= on process() raises TypeError.
Modes¶
| Mode | Description |
|---|---|
balanced |
Default - security + optimization |
strict |
Fail-closed |
lite |
Minimal policy features |
off |
Passthrough |
Performance¶
Token reduction?¶
Typically 5-15% on verbose prompts. Already-short prompts may see little change. See benchmarks.md.
Speed?¶
Roughly 20-80 ms for rule-based PII on typical prompts. Use mode="lite" or mode="off" for lower latency.
Security¶
What PII is detected?¶
Emails, phones, SSNs, credit cards, API keys, JWTs, IPs, and more. See security.md.
GDPR compliant?¶
ASHA is privacy tooling, not a certified compliance product. See compliance.md.
Agent¶
Key parameters?¶
model, privacy, token_budget, provider, routing_config, local_model, sample_prompts.
What does run() return?¶
String by default. AgentResult when trace=True.
Removed in v0.4.1¶
| Removed | Use instead |
|---|---|
Pipeline, Processor |
process() or PromptProcessor |
Root wrap_llm |
asha.integrations.wrap_llm |
security_fail_closed= |
mode="strict" |
return_metrics=True |
result.metrics |
ModelRouter |
Agent(routing_config=...) |
Full list: deprecations.md.