feat(js/plugins/goodmem): add GoodMem plugin

[MohamedKhalil8801]

Closes #5372.

Adds a first-party genkitx-goodmem plugin that exposes GoodMem, a retrieval-augmented generation (RAG) memory service, as Genkit tools.

Contents

• js/plugins/goodmem/: plugin source plus 34 unit tests with mocked fetch.
• js/testapps/goodmem/: runnable demo with three scenarios (persistent project context across sessions, a scribe-and-analyst pipeline using Genkit span enrichment, and a structured team activity log with metadata-driven retrieval).

11 tools are registered under the goodmem/ namespace: list_embedders, list_spaces, get_space, create_space, update_space, delete_space, create_memory, list_memories, retrieve_memories, get_memory, delete_memory.

Usage

import { genkit } from 'genkit';
import { goodmem } from 'genkitx-goodmem';

const ai = genkit({
  plugins: [
    goodmem({
      baseUrl: process.env.GOODMEM_BASE_URL || 'http://localhost:8080',
      apiKey: process.env.GOODMEM_API_KEY!,
    }),
  ],
});

Once loaded, the tools work with any Genkit agent or flow. For example, retrieving memories scoped by metadata:

const retrieve = await ai.registry.lookupAction(
  '/tool/goodmem/retrieve_memories'
);
const { results } = await retrieve({
  query: 'Show me the new features we shipped this cycle.',
  spaceIds: [spaceId],
  metadataFilter: "CAST(val('$.category') AS TEXT) = 'feat'",
});

See js/plugins/goodmem/README.md for the full tool reference and js/testapps/goodmem/src/index.ts for the three-scenario demo.

Verification

• pnpm build in js/plugins/goodmem and js/testapps/goodmem: clean.
• pnpm test in js/plugins/goodmem: 34 / 34 pass.
• Live run of the testapp against a running GoodMem server and OpenAI gpt-4o-mini produces correct grounded answers across all three scenarios.

Checklist (if applicable):

• PR title is following https://www.conventionalcommits.org/en/v1.0.0/
• Tested (manually, unit tested, etc.)
• Docs updated (updated docs or a docs bug required)

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.

[gemini-code-assist]

Code Review

This pull request introduces the genkitx-goodmem plugin, which integrates the GoodMem RAG memory service into Genkit by providing 11 specialized tools for managing spaces and memories. The implementation includes a comprehensive test suite and a demo application showcasing various RAG scenarios. Review feedback recommends replacing synchronous file system operations with asynchronous ones to avoid blocking the event loop and correcting error handling logic that incorrectly attempts to access properties on standard Error objects. Additionally, a discrepancy in the default base URL protocol within the demo application was identified for correction to ensure local connectivity.

[gemini-code-assist]

Using synchronous file system methods like fs.existsSync and fs.readFileSync blocks the Node.js event loop, which can impact performance. It is recommended to use the asynchronous fs.promises API. Additionally, when a tool fails, instead of returning an error object, add a message to the history queue to inform the LLM of the error and return control, allowing the LLM to attempt a correction. Minor optimizations regarding the placement of base64 conversion should be avoided if the current logic is functionally correct.

        if (filePath) {
          try {
            const fileBuffer = await fs.promises.readFile(filePath);
            const ext = path.extname(filePath).replace('.', '');
            const detectedMime = getMimeType(ext);
            const mimeType = detectedMime || 'application/octet-stream';
            fileName = path.basename(filePath);

            requestBody.contentType = mimeType;
            if (mimeType.startsWith('text/')) {
              requestBody.originalContent = fileBuffer.toString('utf-8');
            } else {
              requestBody.originalContentB64 = fileBuffer.toString('base64');
            }
          } catch (e: any) {
            history.push({ role: 'user', content: `Failed to read file at ${filePath}: ${e.message}` });
            return;
          }
        }

References

1. When a tool fails, instead of throwing an error or returning a ToolResponse, add a message to the history queue to inform the LLM of the error and return control.
2. Avoid making changes for minor optimizations if the existing code is functionally correct.

[gemini-code-assist]

The code attempts to access error.response?.body, but the apiJson and apiFetch helpers throw standard Error objects. Furthermore, when a tool fails, instead of returning a failure response or letting the error propagate, add a message to the history queue to inform the LLM of the error and return control, allowing it to attempt a correction.

References

1. When a tool fails, instead of throwing an error or returning a ToolResponse, add a message to the history queue to inform the LLM of the error and return control.

[gemini-code-assist]

The default BASE_URL uses https://localhost:8080, while the plugin's README and typical local GoodMem setups use http://localhost:8080. This discrepancy may cause connection failures for users running the demo locally without a custom environment variable.

Suggested change

	const BASE_URL = process.env.GOODMEM_BASE_URL ?? 'https://localhost:8080';
	const BASE_URL = process.env.GOODMEM_BASE_URL ?? 'http://localhost:8080';

[MichaelDoyle]

Hey @MohamedKhalil8801 - excited to see this come to Genkit, thanks for the contribution!

We're not accepting 3P plugin submissions to the repo at this time. Our recommendation is to publish it to NPM with the prefix genkitx-* (see: https://www.npmjs.com/search?q=genkitx).

[MichaelDoyle]

Closing per my earlier explanation. Thanks again!