netloafer

BladeX — AI skill for Claude Code

AI community

Self-hosted memory-first proxy for AI agents.

How to install BladeX

This entry records only its repository, not the path inside it, so there is no exact command to give. Open netloafer/BladeX and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What BladeX does

Self-hosted memory-first proxy for AI agents. Cross-session & cross-agent memory + task ledger. Zero code changes. Works with Claude Code, Codex CLI, Hermes, Pi, DeepSeek Harness & any OpenAI-compatible client.Your data stays private.

Alternatives in AI

Opencodex — Universal provider proxy for OpenAI Codex & Claude Code — use any LLM (Claude, Gemini, Grok, DeepSeek, Ollama… 12.3k ★
EchoBird — One-click install + model switch:Claude Code,Codex CLI (OpenAI), Grok Build (xAI), DeepSeek Harness, Kimi Code 3.1k ★
WindsurfAPI — Turn Windsurf / Devin Desktop's 100+ AI models (Claude, GPT, Gemini, DeepSeek, Kimi, GLM, SWE) into OpenAI-, A 3k ★

README

**Build your private data assets.** **Connect all your agents and LLMs like a blade.**

BladeX is a **self-hosted, memory-first proxy** that sits between your AI agents and the model providers. Point any agent's API base URL at BladeX and it gains cross-session, **cross-agent** memory plus a per-task ledger — with **no code changes to the agent**. Everything stays on your machine: the memory is a private data asset you own, not a feature you rent.

中文版：[README.zh.md](README.zh.md)

The 4S

	What it means
Save all Memory	Every turn lands in the Memory Hub losslessly — full text and metadata, the single source of truth. Nothing is silently truncated, and nothing is deleted to make room: when capacity is tight, trust decays and facts are archived.
Saving your Token	The injected surface is no longer a memory dump. Each turn carries hard rules, a stable self-description, and the task ledger; everything else the model pulls on demand through `bladex_memory_search`. A ledger that survives compaction also cuts the re-explaining rounds, which is where the real spend is.
Smart LLM Router	One endpoint, many models. Configuration-driven deterministic routing (explicit request model → team → principal → agent), optional LLM judging, session stickiness, health-based failover. Static configuration always beats runtime inference.
Sift out experience	Raw turns are not memory. A background distiller turns them into structured facts, groups them into Matters (a "thing" you work on across sessions and agents), and the model records what it has actually established in the ledger's `Verified` section.