Biological-Silicon
Hybrid Portal
Epichlo Labs synthesizes biological resilience and silicon computation. Architecting high-performance context optimization with TokenDamper and community-owned distributed AI with Public Intelligence.
Biological-Silicon Synthesis
Epichlo Labs operates at the intersection of biological adaptive resilience and raw silicon processing power. We build infrastructure that treats context engineering and distributed intelligence as organic, self-healing networks.
Through high-precision token dampening and decentralized compute topology, our portal empowers developers and researchers to deploy sovereign, resilient intelligence systems without context bloat or infrastructure lock-in.
TokenDamper Engine
Universal context optimization engine for AI coding assistants. Intelligent middleware proxy that compresses and deduplicates context before reaching LLMs — reducing token usage and accelerating response times while guaranteeing 100% semantic correctness.
0/1 Knapsack Planner
Evaluates value-density of AST nodes and file context blocks, optimally packing high-utility context under strict token budgets (--max-input-tokens) while eliminating low-value noise.
Cross-Turn Session Deduplication
Stateful TokenHasher tracks conversation state across turns. Previously transmitted code blocks are replaced with 256-bit SHA-256 hash signatures, cutting redundant tokens by up to 90%.
<BLOCK_HASH> Hashing
Replaces repetitive structural boilerplate (imports, interfaces, headers) with compact <BLOCK_HASH> placeholders during proxy pass, recovering full text losslessly upon model response.
Myers Diff Delta Compression
Computes line and character-level deltas between file snapshots across turns using the deterministic Myers diff algorithm, streaming compact unified patches instead of full files.
Proxy Gateway & MCP Server
Operates in 3 modes: Direct CLI (tokendamper optimize), Transparent Gateway Proxy (tokendamper exec), and Model Context Protocol stdio server (tokendamper mcp) for Claude Desktop & Cursor.
Debt & Drift Safety Validators
Pure validators (ConfidenceLedger, DebtTracker, DriftTracker) enforce zero information loss (--max-debt, --max-drift), triggering automatic fallback to raw input on risk.
TokenDamper CLI Quickstart
Install globally via npm or run as an MCP stdio server to optimize context for Claude Desktop, Cursor, Aider, or custom LLM pipelines.
- [RAW CONTEXT] import { React, useState, useEffect, useCallback, useMemo } from 'react'; // 1,420 tokens
- [RAW CONTEXT] interface UserProfileProps { id: string; name: string; email: string; avatar: string; ... }
+ [DAMPER OPTIMIZED] <BLOCK_HASH: e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855> // 12 tokens
+ [MYERS PATCH] @@ -42,7 +42,7 @@ function renderProfile() { - return <div>Old Profile</div>; + return <div>Updated Profile</div>; }
:: Knapsack Budget: 8,192 / 8,192 tokens filled | SHA-256 Cache Hit: 94.2% | Latency: 1.18ms
- import { React, useState, useEffect, useCallback, useMemo } from 'react';
- interface UserProfileProps { id: string; name: string; email: string; avatar: string; bio: string; }
- export const UserProfileComponent = ({ id, name, email }: UserProfileProps) => {
- const [state, setState] = useState(null);
- useEffect(() => { fetchUserData(id).then(setState); }, [id]);
- return <div>Old Profile Component -- Full AST Tree Payload</div>;
- };
:: Reduction: 0.0% | Raw Token Count: 1,420 tokens | Latency: 0.00ms (Unbuffered)
+ <BLOCK_HASH: e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855> // 12 tokens
+ export const UserProfileComponent = ({ id, name, email }: UserProfileProps) => { ... };
:: Knapsack Budget: 8,192 / 8,192 tokens filled | SHA-256 Cache Hit: 94.2% | Saved: 1,408 tokens (99.1%)
@@ -42,7 +42,7 @@ function renderProfile() {
- return <div>Old Profile Component -- Full AST Tree Payload</div>;
+ return <div>Updated Profile Component -- Myers Compressed Delta Patch</div>;
}
:: Myers Diff Delta: 24 bytes | Compressed Token Count: 18 tokens | Latency: 1.18ms
Public Intelligence Network (Phase 4.5)
Globally distributed, community-owned P2P AI compute mesh powering sovereign LLM inference through Node local workers, Scheduler OpenAI Gateways, and Next.js 16 Visual Control Planes.
Node (Compute Worker)
Executes localized model inference via Ollama inside Docker Sandbox Worktree runtimes. Exposes local telemetry APIs (/api/v1/node/telemetry), start/stop host controls, and live SSE container log streams over Zenoh P2P mesh.
Scheduler (Decision Engine)
Provides an OpenAI-compatible gateway (POST /v1/chat/completions) with RS256 JWT authentication, TokenBucket rate limiting (HTTP 429), real-time SSE token streaming (stream: true), and deterministic node load balancing.
Visual Control Plane
Next.js 16 + React 19 + Tailwind CSS v4 portal featuring /dashboard for host contributors (start/stop node, CPU/RAM/VRAM telemetry, container log viewer) and /playground for interactive requester prompt runs.
One-Click Node Installer (install.sh)
Deploy a Public Intelligence compute worker node in seconds. POSIX installer auto-detects GPU/VRAM hardware, checks prerequisites, sets up WAN P2P networking, and boots the public-intelligence-node daemon launcher.
Request Flow Architecture
Distributed execution lifecycle: Client Request → Scheduler Decision Engine → Compute Worker Node → Ollama Backend
Network & Product Roadmap
Chronological milestone evolution of Public Intelligence and TokenDamper core infrastructure.
v1.0.0 — Autonomous Core
TokenDamper daemon with 0/1 Knapsack context planning, SHA-256 deduplication, Myers diff delta compression, and Public Intelligence Client → Scheduler → Node → Ollama inference execution graph.
v2.0.0 — Decentralized Mesh & ZK Proofs
Decentralized token staking mesh, Zero-Knowledge verification for compute node inference outputs, dynamic knapsack weight adaptation, and global multi-region scheduler consensus.
Engineering Principles
Biological Resilience
Systems adapt and degrade gracefully under network partition or compute failure. Self-healing node routing ensures continuous operational uptime across distributed clusters.
Zero Context Waste
Every token consumed incurs economic and latency cost. Intelligent compression, session deduplication, and knapsack optimization preserve context integrity with minimal overhead.
Open Distributed Consensus
Community-owned infrastructure free from single-point control or proprietary vendor lock-in. Auditable, open execution across decentralized compute workers.