glimind 0.4.0

Live reliability oracle for AI agent tools (MCP) — ask if a tool is up right now before calling it, fail over to a healthy alternative, and report outcomes (privacy-preserving). Zero dependencies.

glimind (Python)

Live reliability oracle for AI agent tools. Ask if a tool works before calling it, and report your own outcomes (privacy-preserving) so the data gets better for everyone. Zero third-party dependencies (stdlib only); reporting is batched on a background daemon thread and flushed at exit, so it adds no latency to your calls.

pip install glimind

from glimind import Glimind

s = Glimind(endpoint="https://glimind.com", api_key="sk_live_...")  # api_key optional

# 1) ask before acting
h = s.check("acme/search")
if h["verdict"] == "down":
    ...  # pick a fallback

# 2) wrap a call — success/failure/latency reported automatically (privacy-clean)
with s.wrap("acme/search", input={"q": query}, task="search"):
    result = acme.search(query)

# or decorate a function tool
@s.instrument("acme/search", task="search")
def search(q): ...

Framework middleware — one line

LangChain / LangGraph:

from glimind.langchain import GlimindCallbackHandler
agent.invoke(input, config={"callbacks": [GlimindCallbackHandler(s)]})

OpenAI Agents SDK:

from glimind.openai_agents import instrument_agent
agent = instrument_agent(s, agent)   # every function tool now auto-reports

Privacy

This client never sends your arguments, payloads, or tool outputs. Before anything leaves your process it is reduced to a value-free shape skeleton (every leaf value becomes a type tag), errors are scrubbed, and only: tool id, input shape, success/failure, normalized error, latency, coarse region are transmitted. The derivation is identical to the TypeScript SDK, so shapes hash consistently across languages.

MIT licensed.