DEDUPED Hot AI repos, turned into use cases Get the Monday 5

How to trace each recalled memory back to the turn it came from with Soda Mem

Check where a stored fact came from before trusting an agent's answer about a user.

Short answer: Run pip install "sodamem[chroma,llm]" to install Soda Mem; then call mem.build_context for the user and question you want to audit, with a token_budget; read block.citations to get each evidence item and its source. All steps.

Who it is forA developer or reviewer who must explain why an agent believes something about a user.

The problemA similarity score alone does not show which conversation produced a remembered claim.

The outcomeEach line of recalled context links to a source span and the exact raw turn, so the claim can be checked.

What you need first

  • Python environment with pip available
  • An LLM provider API key set as SODAMEM_LLM_API_KEY, needed only for ingest
  • Docker or a running daemon is not required for the Python path, but coding tools need `sodamem daemon ensure` first

Steps

  1. Run pip install "sodamem[chroma,llm]" to install Soda Mem.
  2. Call mem.build_context for the user and question you want to audit, with a token_budget.
  3. Read block.citations to get each evidence item and its source.
  4. Use the MCP tool entity_timeline to see one entity's history in order, with each item still pointing at its source.

Prompt for your agent

Using the SodaMem repo at <repo path>, run build_context for user <user_id> with the question <question>. Expected output: a list of citations, each tied to a session and turn. Verify that every line in the context has a citation.

Caveat. Citations trace to stored turns only, so the audit is only as complete as what was ingested.

Use this in your AI agent

Ask your agent

Using the Deduped MCP, get the setup steps for SodaMem/SodaMem and "Trace each recalled memory back to the turn it came from"

Add Deduped to Claude Code once

claude mcp add --transport http deduped https://dedupedai.com/mcp

Cursor, ChatGPT, others

Share: X LinkedIn Hacker News Email Copy link

Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer