Hologres:Integrate Hologres long memory service with DeepSeek Harness

Set environment variables

Run the following commands to set the environment variables:

export MEM0_API_KEY="<YOUR-HOLOGRES-MEM0-API-KEY>"
export DASHSCOPE_API_KEY="<YOUR-DASHSCOPE-API-KEY>"

Parameters:

Write the DSH profile configuration file

In the ~/.dsh/profiles/mem0-holo/ directory, create a file named cordis.patch.yml with the following content:

# Model Studio provider configuration. Replace this block if you use a different LLM provider.
- id: llm-pi-ai
  config:
    providers:
      bailian:
        apiKeyEnv: DASHSCOPE_API_KEY
        api: openai-completions
        baseURL: https://dashscope.aliyuncs.com/compatible-mode/v1
        compat:
          supportsDeveloperRole: false
          maxTokensField: max_tokens
        models:
          - id: deepseek-v3
            input: [text]

# Default model
- id: agent-default-model
  config:
    provider: bailian
    model: deepseek-v3

# Official Mem0 plugin. Point host to the Hologres long memory service endpoint.
- insert:
    - id: mem0
      name: "/absolute/path/to/node_modules/@mem0/deepseek-plugin/dist/index.js"
      config:
        userId: "<YOUR-USER-ID>"
        host: "<YOUR-HOLOGRES-MEM0-ENDPOINT>"

Parameters:

Write a memory

Tell the agent what to remember:

Remember this: our team uses Hologres as its real-time data warehouse, and our main query pattern is vector retrieval combined with graph aggregation.

The agent calls add_memory to write the information to the long memory service.

Recall across sessions

Start a new session and ask the agent:

Which data warehouse does our team use, and what query patterns have we been using?

The agent calls search_memory to retrieve the stored memory and refers to it in the answer.

Manage memories directly

You can also manage memories with natural language:

The search_memory and add_memory tools are not loaded

Check the following items in order:

1. The profile name that you pass at startup matches the profile that you created.

2. cordis.patch.yml exists in the profile directory, and the file that name points to exists.

3. MEM0_API_KEY is set, and its value is prefixed with om-.

A memory search returns no results

Check the following items in order:

1. The long memory service contains memory data.

2. The current userId matches the one used when the memory was written. Memory data is isolated by userId.

3. The environment where DSH runs can reach the endpoint of the long memory service, and its outbound IP address is in the whitelist.

An error occurs when the Model Studio model is called

Check that DASHSCOPE_API_KEY is valid and that the configured model ID, such as deepseek-v3, is available in Alibaba Cloud Model Studio.

Cannot read properties of undefined (reading 'prepare')

Cause: Two copies of the @deepseek-ai dependencies exist, one in the profile and one in the main DSH installation, which makes an internal symbol check fail.

Solution: Follow Step 1 to replace the duplicated dependencies in the profile with symbolic links to the main installation, and then restart DSH.

search_memory failed: Internal Server Error

Cause: When the model calls the search operation with an empty query, the search operation of the long memory service returns HTTP status code 500.

Solution: Modify the plugin file node_modules/@mem0/deepseek-plugin/dist/index.js so that an empty query is routed to the list operation instead.

1. Find the following line:

   query: { type: "string", description: "What to recall.", required: true },

   Replace it with:

   query: { type: "string", description: "What to recall. Optional: if omitted or empty, returns the most recent memories instead of running a semantic search." },

2. Replace the try block of the execute method with the following code:

   async execute({ query, limit, userId: u, agentId, runId }) {
     const filters = resolveSearchFilters({ userId: u, agentId, runId }, userId);
     try {
       const topK = limit && limit > 0 ? limit : DEFAULT_SEARCH_LIMIT;
       if (!query || !String(query).trim()) {
         // For an empty query, call the list operation to return the most recent
         // memories and avoid the error from the search operation.
         const listed = await client.getAll({ filters, pageSize: topK });
         const results = listed?.results ?? [];
         return truncateOutput(formatMemoryList(results));
       }
       const { results } = await client.search(query, { filters, topK });
       return truncateOutput(formatMemoryList(results ?? []));
     } catch (err) {
       return `search_memory failed: ${err instanceof Error ? err.message : String(err)}`;
     }
   }

3. Restart DSH to apply the change.

invalid arguments: missing required property 'query'

Cause: The query parameter is required by default, so the framework rejects the call when the model omits it. DSH usually retries automatically, and the error does not normally affect usage.

Solution: This error has the same root cause as search_memory failed: Internal Server Error. Use the same procedure to make query optional.

The --profile option has no effect

--profile is a global option and must precede the subcommand. For example, to start the web UI, run dsh --profile mem0-holo web instead of dsh web --profile mem0-holo.

MEM0_API_KEY has no effect after it is set

DSH reads environment variables at startup. Export the variable before you start DSH. If DSH is already running, restart it.