Sovereign AI for Everyone!
Built local-first, security-native, owned by you.
software that gives you AI without giving up your sovereignty
Why this exists
Built in the opposite order
In January 2026, OpenClaw hit 60,000 GitHub stars in 72 hours and 300,000+ users in weeks. Kaspersky found 512 vulnerabilities (8 critical). Cisco found active data exfiltration in third-party skills.
LegionForge is built in the opposite order: security first, product on top. Every component lives on your hardware, under your keys. Validation runs in deterministic code, not in the model. The LLM is the last resort, not the first.
Five non-negotiables
Design principles
Fail-safe tiering
Halt → sandbox/retry → degrade. Never silently succeed.
Human gates on mutations
Destructive actions cross a human-in-the-loop boundary by default.
Replace AI with determinism
The LLM is the last resort. Rules, tables, and pattern matchers run ahead.
Validate at trust boundaries
Sanitize once, at the edge. Internal code trusts internal data.
Privilege tied to tasks
Capability is scoped to the active task and expires when it ends.
The ecosystem
Build with what fits
LegionForge is a family of independent, composable projects. Adopt the framework end-to-end, or use Guardian standalone to harden the agent stack you already run.
LegionForge Framework
Local-first AI agent framework, security in the foundation
coreGuardian
Deterministic security sidecar for any agent framework
appJeli
Sovereign, cryptographically-attested personal memory
appADHD-OS
Personal assistant framework for those with ADHD
toolllm-valet
Auto-pause/resume Ollama based on system pressure
toolmcp-probe
Connectivity and configuration advisor for your MCP services
toolheadroom
System stability monitor with AI-powered diagnostics
toolhermes-tool-test-suite
pytest harness for AI agent tool-calling reliability
tooldev-rig
Shared CI workflows and audit harness for LegionForge projects
Security posture
Different from cloud agent platforms — different from unguarded OSS frameworks
Most cloud agent platforms (OpenAI Operator, Anthropic Computer Use, Google Mariner) run your tasks on someone else's hardware. Your prompts, your data, your tool outputs — they pass through systems you can't audit.
Most open-source frameworks (LangChain, AutoGen, CrewAI) are flexible substrates that let you add guardrails. LegionForge enforces them — in deterministic code, on every step, with no opt-out path.