Agent Reliability Python Packages

superlocalmemory

World's first local-only AI memory to break 74% retrieval and 60% zero-LLM on LoCoMo. No cloud, no APIs, no data leaves your machine. Additionally, mode C (LLM/Cloud) - 87.7% LoCoMo. Research-backed. arXiv: 2603.14588

dunetrace

Real-time monitoring of production AI agents. No raw content transmitted.

kalibr

Stop overpaying to run your agents. Kalibr routes every request to lower-cost model and tool paths without degrading performance.

sourcerykit

SourceryKit — counterspell for hallucinating agents. Python SDK that breaks the illusion on every tool call, API response, and MCP handoff before bad outputs propagate.

auditable

Audit any agent decision across its past, present, and future, on one typed graph.

misfire

misfire — a trace-grounded auditor that finds which CLAUDE.md rules your agents actually ignore and turns the ones that matter into deterministic hooks.

agentassert-abc

Formal behavioral specification and runtime enforcement for autonomous AI agents. Agent Behavioral Contracts (ABC).

self-heal-llm

Automatic repair for failing Python code, powered by any LLM.

aeonik-ingrain

Learned experience layer for autonomous agents.

warmwinter

Python SDK for the Warm Winter trust gate — is this AI/agent action trustworthy enough to act on, or escalate?

harness0

Layer 0 of agent reliability — the framework-agnostic harness that makes any AI agent stable.

llm-cost-optimizer

Automatically reduce LLM API costs. Routes to cheapest model that succeeds. Use kalibr instead.

llm-cost-reduction

Stop overpaying to run your agents. Kalibr routes every request to lower-cost model and tool paths without degrading performance.

llama-index-storage-chat-store-superlocalmemory

LlamaIndex chat store backed by SuperLocalMemory V2 — 100% local, zero cloud

self-improving-loop

A safety-first self-improvement loop for AI agents: execute → track → analyze → auto-apply → auto-rollback on regression.