TetherMesh ⚡

Zero-config local desktop control plane and proxy gateway (127.0.0.1:4000) for autonomous AI coding agents.
Provides automatic model failovers, hard runaway spend caps, 1-click dynamic MCP tool syncing across 6+ client environments, live token telemetry HUD, and full-spectrum activity tracing.

⚡ 60-Second Quickstart Guide

Get up and running with TetherMesh in 3 effortless steps:

1. Download & Launch TetherMesh

• Run the TetherMesh desktop application (.exe, .dmg, or .deb).

• TetherMesh automatically spins up the local proxy gateway on http://127.0.0.1:4000.

2. Enter Your Provider API Keys

• Click "Keys" or open the 60-Second Quickstart Wizard in the top bar.

• Add your Anthropic, OpenAI, AWS Bedrock, Google Vertex, Groq, or Local Ollama credentials.
  (Keys remain strictly local in your desktop vault and are never sent to third-party cloud servers).

3. Connect Your Coding Agent / Tool

Choose your agent below and connect in one step:

🟢 Claude Code CLI

export ANTHROPIC_BASE_URL=http://127.0.0.1:4000
claude

Or click "Auto-Configure Claude Code" in TetherMesh to inject it into ~/.claude.json.

🟣 AI IDEs (Cursor, Windsurf, Devin, Antigravity)

• Base URL: http://127.0.0.1:4000/v1

• Model: fast-code (low-latency) or heavy-reasoning (deep reasoning)

• MCP Auto-Sync: Click "Connect & Auto-Configure All Files" in TetherMesh's Tool Hub to inject any of the 50+ MCP servers into ~/.cursor/mcp.json, ~/.codeium/windsurf/mcp_config.json, devin.json, or .mcp.json.

🔵 Python & TypeScript SDKs

from openai import OpenAI

client = OpenAI(
    base_url="http://127.0.0.1:4000/v1",
    api_key="tethermesh-local"
)

response = client.chat.completions.create(
    model="heavy-reasoning",
    messages=[{"role": "user", "content": "Refactor this architecture"}]
)
print(response.choices[0].message.content)

Related MCP server: Bastion

🛡️ Core Features

1. Dual Gateway Engine (127.0.0.1:4000):

   • Full OpenAI (/v1/chat/completions, /v1/models) and Anthropic (/v1/messages) protocol compatibility.

   • Low loopback latency ($< 5\text{ ms}$).

2. Resilient Model Failover Matrix:

   • Linear and tiered priority chains (e.g. Anthropic $\rightarrow$ AWS Bedrock $\rightarrow$ Groq $\rightarrow$ Ollama).

   • Silent 429 Too Many Requests and 503 Service Unavailable recovery without breaking active agent CLI sessions.

   • Virtual Model Aliases: fast-code and heavy-reasoning.

3. Hard Runaway Spend Circuit Breakers:

   • Visual daily and monthly spend limit dials (e.g. $10.00/day).

   • Automatically short-circuits runaway recursive agent loops with HTTP 402 Budget Exceeded: TetherMesh spend limit reached.

4. 50+ Official & Verified MCP Tool Marketplace:

   • Verified schemas for Databricks, Snowflake, Supabase, PostgreSQL, Notion, Slack, Jira, GitHub, Docker, Sentry, Brave Search, Pinecone, and 40+ more.

   • 1-Click simultaneous non-destructive configuration file injection across Cursor, Windsurf, Devin, Claude Code CLI, Claude Desktop, and Antigravity.

5. Deep Observability & Activity Tracing:

   • Real-time span waterfall visualizer.

   • Microsecond latency breakdowns, prompt/payload inspectors, and OpenTelemetry-compatible traces.

6. Embedded Execution Terminal:

   • Integrated drawer running native host shells (powershell.exe, zsh, bash) with pre-loaded proxy gateway environment variables.

7. 1-Click Sanitized Debug Reporter:

   • Generates clean GitHub issue markdown with all API keys and bearer tokens automatically redacted (sk-ant-***, ghp_***).

🛠️ Development & Building Locally

# Install dependencies
npm install

# Start Vite React UI
npm run dev

# (Option A) Run LiteLLM Proxy in Development
litellm --port 4000 --host 127.0.0.1 --config sidecar/dev_config.yaml

# (Option B) Build LiteLLM Standalone Sidecar Binary (Windows)
npm run sidecar:build

# Build production frontend bundle
npm run build

Built with Tauri v2, React 19, TypeScript, Tailwind CSS, and official LiteLLM Proxy Core.