LLMFoundry v1.0.0
The AI engineering kit for DeepSeek. Turn opencode into a team: an orchestrator that
understands your intent, five specialist agents, a living memory that learns and feeds
itself, and gates that protect your work.
What's inside
- 29 skills across 5 categories: dev process (SPEC, TDD, git, PR, code review),
AI core (prompt engineering, agents, RAG, evals, MCP, observability), advanced
(research with correlation, doubt-driven development), and a full reverse engineering
cluster (binary analysis, decompilation, algorithm recovery, malware, firmware). - 6 agents: the AI Orchestrator (default) plus deep-researcher, ai-architect,
ai-evals-runner, llm-security-reviewer, and reverse-engineer. - 7 slash commands:
/ai-spec,/ai-build,/ai-evals,/ai-review,
/ai-research,/ai-memory,/ai-re. - 4 plugins: gates (block bad commits/egress), memory (living loop + recall),
voice-guard (never look AI-generated), verify-guard (anti-delirium).
Disciplines
- Anti-delirium: every claim carries concrete proof or an honest confidence marker.
- Human voice: output never reads like AI-generated text.
- Cost discipline: DeepSeek only, with explicit PRO vs FLASH routing.
- Mode discipline: PLAN before BUILD when there is ambiguity or stakes.
- Privacy by design: memory is 100% local, never versioned.
Memory
SQLite + FTS5 + local semantic embeddings. A living feedback loop that agents feed and
consume: encode, consolidate, retrieve, reconsolidate. Auto-captures errors and findings,
injects recall into every session.
Quality
- 32 deterministic checks (engine, routing, plugins, K=5 stability), no model calls.
- GitHub Actions CI runs them on every push.
- Chrome DevTools MCP connects to your Chrome with your logins via autoConnect.
Install
git clone git@github.com:Pl3ntz/llmfoundry.git
cd llmfoundry
./scripts/install.shRequires: opencode, python3, DeepSeek V4 (Go plan or API key).