Skip to main content

How to verify memory is used

This guide shows that Talon loads stored memory and injects it into the prompt before calling the LLM. Same agent and tenant for all commands (memory is scoped by tenant_id and agent_id).

Prerequisites​

  • Talon built and an LLM API key (e.g. export OPENAI_API_KEY=sk-...)
  • Commands run from the same directory; agent name comes from agent.name in your policy

Steps​

1. Create a test project with memory enabled

mkdir -p /tmp/talon-memory-demo && cd /tmp/talon-memory-demo

Use a policy with memory.enabled: true and optional max_prompt_tokens for relevance retrieval. Example minimal block:

memory:
  enabled: true
  mode: active
  max_entries: 50
  max_prompt_tokens: 500
  allowed_categories:
    - domain_knowledge
    - factual_corrections
  governance:
    conflict_resolution: auto

2. Teach something

talon run "Remember: our company headquarters is in Berlin."

You should see a normal run and evidence stored. The runner writes a memory entry from the response.

3. Confirm the entry exists

talon memory list

Use the same agent as the run (omit --agent to use the policy’s agent.name, or pass it explicitly).

4. Ask something that needs that memory

talon run "Where is our company headquarters?"

If memory is used, the runner injects a [AGENT MEMORY INDEX] block into the prompt. The reply should reflect the stored fact (e.g. Berlin).

5. Prove it in the audit trail

talon audit show

When memory was injected, the evidence shows Memory reads and Memory tokens.

Optional​

  • No memory for one run: talon run --no-memory "Where is our HQ?" — no memory read or write; useful to compare behavior.
  • Point-in-time (compliance): talon memory as-of <RFC3339> --agent <name> — entries valid at that time.

See Memory governance for configuration, modes, and compliance.

You're done​

You now know how to confirm that Talon loads and injects memory into the prompt. Evidence shows memory reads and tokens when memory was used.

Next steps:

I want to…Doc
Configure memory categories and retentionMemory governance
Run an agent with tools or triggersYour first agent with Talon
Export evidence that includes memoryHow to export evidence for auditors
Understand the request lifecycleWhat Talon does to your request