How do I run several agent memory lanes or change the fallback key for remembered observations?

Tares needs no setup for agent memory, because the first remember call auto-provisions an agent_memory source. You create a memory source yourself, in the console or in YAML, only when you want a custom fallback key or several memory lanes. The key setting is the fallback entity key used when a memory’s payload has none, and it defaults to agent.

How it works

The memory connector stores an agent’s own observations. Agents write them with the MCP remember tool, or with POST /remember. Every connected agent has the tool.

• The first remember call auto-provisions a source named agent_memory. You do not add anything beforehand.
• Each observation is keyed by the entity it was remembered against, for example service=checkout.
• When a memory’s payload has no entity key, Tares uses the source’s fallback key. That is the key setting, and its default is agent.
• Memories join reads and triggers like any other source. What an agent concluded about service=checkout last week sits in the same timeline as that service’s logs and deploys.

The connector has one configuration field:

To see the full connector reference, open Agent memory: memory. For the wider picture of how sources feed investigations, see AI SRE.

Set it up

Start with the default. Ask a connected agent to remember something, and Tares creates the agent_memory source for you:

Use tares to remember that the checkout latency spike on July 1 was caused by the pool-size change.

Create a memory source explicitly only when you need a custom fallback key or several lanes.

1. Decide what each lane is for. Each memory source is one lane, so give each its own name.
2. Add a memory source in the console, or declare it in YAML. A source with the default name and key looks like this:

- name: agent_memory
  connector: memory
  config:
    key: agent

3. To change the fallback key, set key to the value you want. Tares uses it whenever a memory’s payload has no entity key.
4. To run several lanes, add one memory source for each lane, each with its own name and its own key.
5. Verify. After a remember call, the observation appears in the source’s Recent events, keyed by the entity it was remembered against. It also appears in that entity’s correlated timeline.

What you get

• Agent memory that works from the first remember call, with no configuration.
• A fallback key you control, so memories without an entity in their payload land under a key that makes sense for your team instead of agent.
• Separate memory lanes, each its own source, when you want to keep different kinds of observations apart.
• Observations that sit in the same correlated timeline as logs, deploys and other sources, so a later investigation sees what an agent concluded earlier.
• A Recent events view on each memory source, where you can confirm what was written.

When something else fits

• If you only need one lane and the key agent is fine, skip the explicit source and let the first remember call provision it.
• If you want agents to record structured conclusions for people to review, rather than free-form observations, see the delivery options in AI SRE.

Related