Convex CommunityCCConvex Community
Powered by
David AlonsoD
Convex Community•2y ago•
5 replies
David Alonso

Passing `ctx` to AI agent tools

I'm currently doing the following inside a convex action:
const result = await createBlockFromPromptAls.run(
      {
        ctx: ctx,
        clerkOrgId: args.clerkOrgId,
        blockId: args.blockId,
        workspaceCollectionInfo: args.workspaceCollectionInfo,
      },
      () =>
        generateText({
          model: anthropic("claude-3-5-sonnet-20240620"),
          prompt,
          maxToolRoundtrips: 5, // allow up to 5 tool roundtrips
          tools: {
            getFields: FireviewAITools.getFields,
            createTableBlockOneShot: FireviewAITools.createTableBlockOneShot,
          },
        })
    );
const result = await createBlockFromPromptAls.run(
      {
        ctx: ctx,
        clerkOrgId: args.clerkOrgId,
        blockId: args.blockId,
        workspaceCollectionInfo: args.workspaceCollectionInfo,
      },
      () =>
        generateText({
          model: anthropic("claude-3-5-sonnet-20240620"),
          prompt,
          maxToolRoundtrips: 5, // allow up to 5 tool roundtrips
          tools: {
            getFields: FireviewAITools.getFields,
            createTableBlockOneShot: FireviewAITools.createTableBlockOneShot,
          },
        })
    );


where I'm using an ALS like so:
export const createBlockFromPromptAls = new AsyncLocalStorage<{
  ctx: AuthActionCtx;
  clerkOrgId: string;
  blockId: Id<"blocks">;
  workspaceCollectionInfo: z.infer<typeof zWorkspaceCollectionsInfo>;
}>();
export const createBlockFromPromptAls = new AsyncLocalStorage<{
  ctx: AuthActionCtx;
  clerkOrgId: string;
  blockId: Id<"blocks">;
  workspaceCollectionInfo: z.infer<typeof zWorkspaceCollectionsInfo>;
}>();


so that tools can access the ctx to query or write to the Convex db, e.g.

export const getFields = tool({
    parameters: ...
    execute: async ({ collectionId }) => {
      const store = createBlockFromPromptAls.getStore();
      const { ctx, clerkOrgId } = store;
      ctx.runQuery...
export const getFields = tool({
    parameters: ...
    execute: async ({ collectionId }) => {
      const store = createBlockFromPromptAls.getStore();
      const { ctx, clerkOrgId } = store;
      ctx.runQuery...


HOWEVER, I'm running into:
[ERROR] Could not resolve "node:async_hooks"
[CONVEX]     lib/ai/als.ts:1:34:
[CONVEX]       1 │ import { AsyncLocalStorage } from "node:async_hooks";
[CONVEX]         ╵                                   ~~~~~~~~~~~~~~~~~~
[CONVEX] 
[CONVEX]   The package "node:async_hooks" wasn't found on the file system but is built into node. Are you trying to bundle for node? You can use "platform: 'node'" to do that, which will remove this error.
[ERROR] Could not resolve "node:async_hooks"
[CONVEX]     lib/ai/als.ts:1:34:
[CONVEX]       1 │ import { AsyncLocalStorage } from "node:async_hooks";
[CONVEX]         ╵                                   ~~~~~~~~~~~~~~~~~~
[CONVEX] 
[CONVEX]   The package "node:async_hooks" wasn't found on the file system but is built into node. Are you trying to bundle for node? You can use "platform: 'node'" to do that, which will remove this error.


which i could fix by using the
use node
use node
directive, but I'm trying to stay away from it for performance reasons. Any ideas on how do do this with edge runtime packages?
Convex Community banner
Convex CommunityJoin
Join the Convex Discord! Explore Convex, the backend platform that keeps your app in sync.
15,010Members
Resources
Recent Announcements

Similar Threads

Was this page helpful?
Recent Announcements
james

Hi @everyone with everything going on today I forgot to post on discord about our EU launch! We have a region in Dublin now! You can start using it immediately, even for folks on the free plan. https://news.convex.dev/we-finally-got-our-eu-visa/ Stay tuned for more regions but also stay tuned for more infra improvements where we improve latency for teams no matter where their servers are located. Just at the tip of the iceberg re all the optimizations we have lined up.

james · this hour

Wayne

Hi @everyone happy Monday. Components Authoring [Challenge](https://www.convex.dev/components/challenge) updates! Meet the second batch of community-approved components **Firecrawl Scrape **- Scrape any URL and get clean markdown, HTML, screenshots, or structured JSON - with durable caching and reactive queries. https://www.convex.dev/components/firecrawl-scrape Built by: Gitmaxd **Durable Agents **- A Convex component for building durable AI agents with an async tool loop. https://www.convex.dev/components/durable-agents Built by: Siegfried **Convex Debouncer** - A server-side debouncing component for debouncing expensive operations like LLM calls, metrics computation, or any heavy processing that should only run after a period of inactivity. https://www.convex.dev/components/debouncer Built by: Ilya **DatabaseChat **- A Convex component for adding natural language database queries to your app. https://www.convex.dev/components/database-chat Built by: Nick **Transloadit** - A Convex component for creating Transloadit Assemblies, handling resumable uploads with status, and persisting status/results in Convex. https://www.convex.dev/components/transloadit Built by: Kvz **Loops** - A Convex component for integrating with Loops.so email marketing platform. https://www.convex.dev/components/loops Built by: Bobby The [challenge](https://www.convex.dev/components/challenge) is now ongoing, so keep building, and we'll keep rewarding. Thanks, everyone!

Wayne · 2w ago

Liz C

Hi everyone! Have you ever wanted to get your hands on some convex swag? Well you're in luck! We just launched our Convex swag store. Check it out here ---> https://store.convex.dev/

Liz C · 3w ago

Similar Threads

AI Agent
ZerodayZZeroday / support-community
10mo ago
Issue with AI agent
Bacio001BBacio001 / support-community
8mo ago
Convex ai-agent with Gemini?
Ai-fyAAi-fy / support-community
6mo ago
Upgrading convex-dev/agent to AI SDK v5 :)
ZaidZZaid / support-community
7mo ago