Agent Hub - Claude Code and Codex Optimizer

Make Claude Code and Codex write better code with fewer tokens.

Agent Hub optimizes repo context, routes each task by success-per-token, validates outputs, and retries failed attempts automatically.

Product Proof

Claude Code + Agent Hub

Run Agent Hub: Run Personal Benchmark to generate your local savings report, or use the CLI:

agent-hub benchmark --dataset before-after-10 --baseline claude-sonnet --route coding --export claude.json
agent-hub benchmark compare claude.json --dataset before-after-10

Token savings are measured as raw agent request tokens versus the optimized Agent Hub request tokens actually sent, not as repository size minus compressed repository size.

One-Click Onboarding

Start with one button:

• Connect Claude Code
• Connect Codex
• Boost + Save Tokens

Then switch Boost Mode from Balanced to Save Tokens, Best Code, Fast Fix, Big Refactor, or Local First when the task needs it.

Supports OpenAI, Claude, Gemini, Ollama, OpenRouter, Codex CLI, Cline, Roo Code, Continue, LM Studio, and local models through a single OpenAI-compatible endpoint.

Benchmark Your Agents Locally

Unlike most AI tools, Agent Hub includes a reproducible benchmark corpus for raw-agent comparisons.

You can:

• Compare Claude Code alone vs Claude Code + Agent Hub
• Compare Codex alone vs Codex + Agent Hub
• Compare Cline alone vs Cline + Agent Hub
• Measure tokens saved, task success, retries, patch quality, files touched, time to useful output, and cost saved
• Replay historical decisions and verify why each route won

No vendor benchmarks required.

Use before-after-10 for Marketplace screenshots and coding-100 for the larger reproducible coding proof set.

Explain Every Decision

Each task explanation records the evidence behind the optimization:

• files selected from the repository
• tokens saved by context compression
• model selected by success-per-token
• quality checks and retry strategy
• rejected candidates and failover events

Inspect exactly why a model was chosen and why alternatives were rejected with Agent Hub: Explain Route or agent-hub replay-route <request-id>.

What Agent Hub Optimizes

Cost

Route simple tasks to cheaper models.

Reliability

Automatically fail over when providers are unavailable.

Performance

Choose models based on repository characteristics and task complexity.

Transparency

Inspect routing decisions instead of trusting a black box.

Compatibility

Use multiple coding agents and AI providers through a single endpoint.

Competitive Differentiation

Screenshots

Benchmark Proof

Verify token savings, quality lift, retries, and cost using reproducible before/after benchmarks.

Route Replay

Inspect every routing decision and rejected alternative.

Learning Dashboard

See how routing performance evolves over time.

Provider Health

Monitor latency, availability, quotas, and failovers.

Agent Hub Dashboard

Unified control center for orchestration and optimization.

Setup Screens

Connect coding agents after the value and proof are established.

Common Use Cases

Cline + Multiple Providers

Automatically route requests between Claude, OpenAI, Gemini, and local models.

Cost Reduction

Use smaller models when they are likely to succeed and reserve premium models for difficult tasks.

Local-First Development

Prioritize Ollama or LM Studio and fall back to cloud providers when necessary.

Team Standardization

Expose a single endpoint while allowing multiple underlying providers.

AI Benchmarking

Measure model performance on your own tasks instead of relying on public leaderboards.

Demo

If the Marketplace does not autoplay GIFs in your client, open the screenshots above for the same workflows.

Install

1. Install the extension.
2. Install Python 3.11 or newer.
3. Open a project folder.
4. Click the Agent Hub icon.
5. Click Start.

The Agent Hub backend is bundled with the VSIX.

Connect A Coding Agent

Use Agent Hub as an OpenAI-compatible endpoint from Cline, Roo Code, Continue, Claude Code, or another coding agent:

Base URL: http://127.0.0.1:8787/v1
API Key: local-agent-hub-token
Model: agent-hub-coding

Use Agent Hub: Copy Cline Config and Agent Hub: Test Cline Connection from the command palette to verify setup before a real task.

Provider Setup

Pick one or more provider paths:

• Save an OpenAI, Claude, Gemini, Groq, OpenRouter, or other API key in Settings.
• Start Ollama or LM Studio locally and let Agent Hub route to the local endpoint.
• Run Agent Hub: Install Ollama Desktop if the ollama command is not present.
• Run Agent Hub: Install Codex CLI, sign in, then use Agent Hub: Use Signed-In Codex CLI.
• Run Agent Hub: Use Free Models Only to keep only free, local, or free-tier routes eligible.

Ollama Setup

1. Install Ollama Desktop from https://ollama.com/download, or run Agent Hub: Install Ollama Desktop.
2. Pull a coding model, for example ollama pull qwen2.5-coder:7b.
3. Start the Ollama app or run ollama serve.
4. Click Start in Agent Hub, then open Health to confirm the local provider is reachable.

Local routes use http://127.0.0.1:11434. Hosted Ollama Cloud entries require OLLAMA_API_KEY and are treated like other cloud providers.

Safety

Agent Hub asks before sensitive file, shell, process, and provider actions. Dangerous shell commands, path escapes, secret-like payloads, and unknown external endpoints are blocked or routed through explicit approval.

Troubleshooting

More Help

• Cline setup
• Claude Code setup
• Continue setup
• Permissions
• Troubleshooting