MastraM
Mastra•6mo ago
Ata Sanchez

Memory and tool issues.

We constantly have problems when enabling memory and working with tools. The agent loses the thread it's working on and re-executes tools that had already been executed. Previously, we had lastMessages: 10, and it was even more unstable.

Is there a way to "purge" this type of behavior, or what strategy would be best? When a message is sent to the model, it usually freaks out and re-executes a tool it previously executed, causing a tool to be executed multiple times.

Our solution, for now, is to clear the memory on demand when it starts freaking out and send it the normal message. In this case, it works perfectly.

Any ideas about what could be happening?

Memory configuration when use agent.stream

resourceId: userId,
        threadId: `quoter-chat-${chatId}`,
        runtimeContext,
        temperature: 0.3,
        maxSteps: 15,
        memoryOptions: {
            semanticRecall: false,
            lastMessages: 5,
            workingMemory: {
                enabled: false,
            },
        },


Memory configuration:
import { Memory } from '@mastra/memory';
import { ToolCallFilter } from '@mastra/memory/processors';
import { PgVector, PostgresStore } from '@mastra/pg';
import { models } from './models';

if (!process.env.DATABASE_URL) {
    throw new Error('DATABASE_URL must be set');
}

export const memory = new Memory({
    storage: new PostgresStore({
        connectionString: process.env.DATABASE_URL,
    }),
    vector: new PgVector({ connectionString: process.env.DATABASE_URL }),
    embedder: models.textEmbeddingModel('embedding-model'),
    processors: [new ToolCallFilter()],
});
Was this page helpful?