FiberAgent Runtime Harness

One local daemon between AI agents and the tools and models they call.

Local MCP Gateway

Routes your existing MCP servers through one loopback endpoint without rewriting IDE configs. Retried calls return the original result instead of running twice, and risky calls wait for approval via Slack, CLI, or TOTP before any I/O.

Learn more

🛡️

LLM Guardrails

Masks PII and blocks leaked credentials (33 built-in patterns) before requests reach a provider, and cuts off streams that exceed budget. The same rules apply to every provider, over HTTP or in-process. API keys stay on the host.

Learn more

📼

Deterministic Replay

Records LLM streams and tool calls to local JSON fixtures and replays them in pytest with no network or API cost, including streams cut off mid-flight by policy.

Learn more

🌐

Edge Plane (Optional)

Run WASM and Pyodide workloads on Cloudflare Workers you deploy yourself. Quotas, policies, and the state ledger stay on your machine.

Learn more

Quick Start

bash

$ brew install dphilabs/fiber/fiber
# or: uv pip install git+https://github.com/dphilabs/fiber.git@v1.1.5.0

$ fiber init          # create ~/.anchor and node keys
$ fiber connect all   # import IDE MCP configs (originals are not modified)
$ fiber console       # start the gateway and Console Studio

Then point your agent at the gateway:

json

{ "mcpServers": { "fiber-gateway": { "url": "http://127.0.0.1:8356/v1/mcp-gateway/invoke" } } }

FiberAgent Runtime Harness

Local MCP Gateway

LLM Guardrails

Deterministic Replay

Edge Plane (Optional)

Quick Start ​

Quick Start