Om prakash prajapat

Git tracks your source code. VectorForce tracks what your AI coding agent was thinking. Every autonomous coding agent today—Google Antigravity, Meta Muse, Claude Code, Cursor, Windsurf, Copilot, and OpenAI Codex—operates in ephemeral, disconnected chat sessions. When an agent context window closes or when a developer switches tools, three critical failures occur: 1. Architectural Decisions (ADRs) vanish: Non-negotiable security boundaries, API contracts, and design rationale are completely forgotten. 2. Silent Workspace Drift goes unnoticed: Files and Git commits mutate behind the agent’s back without cryptographic validation. 3. Multi-Agent Swarms break: Autonomous agents cannot cleanly hand off work across different IDEs and machines without context degradation. We call this crisis Coding Agent Amnesia. VectorForce is the open-source, local-first engineering memory platform built to solve it. KEY ARCHITECTURAL INNOVATIONS: 1. Cryptographic Claims vs. Evidence Fencing VectorForce introduces a strict boundary between unverifiable human/model claims (goals, architectural decisions, and next steps) and immutable ground-truth evidence (verifiable Git commit SHAs, dirty working-tree states, and SHA-256 document digests). Hallucinated AI claims can never corrupt verified workspace truth. 2. Zero-Dependency Local Core Built with zero external runtime dependencies on Node.js >= 22.5.0 and native SQLite WAL (node:sqlite, node:crypto, node:http). Instant 2-second startup, zero cloud lock-in, and zero data leakage. 3. 8-Harness MCP Auto-Configuration With a single command—"vf config install all"—VectorForce automatically generates native Model Context Protocol (MCP) bridges for Google Antigravity, Meta Muse, Claude Code, OpenAI Codex, Cursor, Windsurf, GitHub Copilot, and Kiro IDE. Agents automatically discover active tasks, detect drift, and commit immutable provenance checkpoints without manual JSON editing. 4. Live Workspace Drift Auditing Before resuming any task, VectorForce cryptographically reconciles current working-tree files against the last committed checkpoint digest, alerting agents immediately if files were tampered with. 5. AI Engineering Memory & Reasoning Layer Using Meta Muse Spark and Llama 3.3 models, "vf reason" synthesizes checkpoint lineage to produce architectural risk warnings, context continuations, and ranked next actions, cleanly degrading to deterministic rule-based briefings when offline. 6. VectorForce Studio & Zero-Knowledge Cloud Includes an interactive local React + XYFlow architecture topology visualizer on port 3210, plus an E2EE cloud sync layer using X25519 ECDH and Ed25519 digital signatures on Modal.com. Validated with a 5-agent banking swarm, 94 automated tests, and available globally on npm: npx vectorforce init.