@fiberplane/agents v0.5.3
@fiberplane/agents
A toolkit for introspecting and debugging agents built with Cloudflare's Agents SDK.
Overview
@fiberplane/agents
provides debugging and introspection capabilities for agents built with Cloudflare's Agents SDK. It allows you to monitor agent state, track messages, and view instance information in real-time.
Installation
npm install @fiberplane/agents
# or
yarn add @fiberplane/agents
# or
pnpm add @fiberplane/agents
Usage
1. Wrap your worker's fetch entrypoint
Use the fiberplane
wrapper for your worker's fetch entrypoint:
import { fiberplane } from "@fiberplane/agents";
export default {
fetch: fiberplane(
async (request: Request, env: Env, ctx: ExecutionContext) => {
// Your existing worker logic...
return await routeAgentRequest(request, env) ||
new Response("Not found", { status: 404 });
}
),
};
2. Use the withInstrumentation
Add the withInstrumentation
mixin to any agent class you want to introspect:
import { withInstrumentation } from "@fiberplane/agents";
import { Agent } from "agents";
class SpecialAgent extends Agent<MyEnv, MyState> {
// Your agent implementation...
}
const MySpecialAgent = withInstrumentation(SpecialAgent)
2b. Bonus step: gain visibility into your LLM calls
When you use Cloudflare's AI Gateway you can gain visibility into the most recent calls by adding an CLOUDFLARE_API_TOKEN
and CLOUDFLARE_ACCOUNT_ID
to your .dev.vars
file. The API token only needs AI Gateway:Read
permissions. See the Cloudflare documentation around Create API Token
3. Access the Agents playground in your browser by appending /fp
to your worker's URL
Features
- State Introspection: Monitor the internal state of your agents
- Message Tracking: View incoming and outgoing messages between agents
- Instance Monitoring: Track each instance of your agents
- Database Inspection: View the state of underlying SQLite database used by the agent
- AI Gateway Inspection: Gain visibility into the calls that go through Cloudflare's AI Gateway
Current Status
This package is in early development and is considered a work in progress. APIs may change in future releases. Also, please do not use this package in production yet.
Publishing a New Version
This package uses the changeset
workflow for versioning:
- Run
pnpm build:agents
from the root of the monorepo - Run
pnpm changeset
from the root of the monorepo - Select major/minor/patch version increment
- Add a summary of changes
- Run
pnpm changeset version
when happy with the change description - Commit changelog changes
- Run
pnpm -F @fiberplane/agents publish
License
MIT/Apache-2.0
6 months ago
6 months ago
6 months ago
6 months ago
6 months ago
6 months ago
6 months ago
6 months ago
6 months ago
6 months ago
6 months ago
6 months ago
7 months ago
7 months ago
7 months ago
7 months ago
7 months ago
7 months ago
7 months ago
7 months ago
7 months ago
7 months ago
7 months ago