Keep background work inspectable.
Some work lasts longer than one tool call. Memrain has a durable job queue and optional research jobs, with status that a connected client can inspect.
Illustrative example
Can I check whether the transcript import finished?
Read the job’s status and progress, inspect any reported failure and retry or cancel through the supported operations.
Illustrative scenario, not a live product result.Make long-running work visible.
Supported server jobs are stored in Postgres so their records survive a process restart. Tools expose status, progress, results and logs, with cancellation and retry operations. The queue runs job kinds registered by the server. It is not a place to submit arbitrary shell commands, and cancellation does not imply that every already-completed side effect can be reversed.
Enable bounded research deliberately.
The optional submit_agent tool queues a read-only research agent that uses the caller’s grants, sources and daily budget. It requires the agent scope, an applicable budget and the server feature flags. The result can be retrieved through get_agent_job. This is a narrower capability than a general autonomous agent platform: it does not promise unrestricted execution, external app actions or an independent digital employee.
Keep orchestration responsibilities clear.
The bundled skills distinguish durable server jobs from the connected agent’s own subagents. Recurring work uses host or harness scheduling, while the maintenance cycle already handles routine memory upkeep. Mid-flight steering and pause-and-resume should not be assumed. When the task changes, inspect the current state and use cancellation or a new bounded task according to the supported workflow.
Keep the context. Move the work forward.
Connect your knowledge to the AI you already use. Find your starting point and make the next conversation count.