Injection: decisions in context
Injection is the most direct answer to “confident is not correct.” Instead of hoping the agent remembers a decision, or re-explaining it every session, Meetless Agent puts the confirmed decisions into the agent’s context at the moment it starts working.
What happens on every prompt
When you submit a prompt, the connector reads the confirmed, trusted slice of governed memory for this workspace and adds it to the agent’s context. The agent now sees your team’s rules alongside your request, before it writes a line.
The effect is mundane and powerful: you stop re-litigating settled decisions. The constraint that “we use Postgres, not Redis, for durable state” is simply present every time, so the agent does not helpfully reintroduce Redis on turn forty.
Client-side, and it does not depend on a model
Injection is deliberately boring where it counts:
- It runs client-side. The confirmed rules are read from local state that the connector maintains for this workspace.
- It does not require a network call or an LLM call to happen. Placing your decisions in context is a local read, not an inference. That makes it fast and keeps it working regardless of what any remote service is doing.
Preview before you trust it
You should not have to take injection on faith. You can inspect exactly what will be placed in context before you rely on it, so injection is auditable rather than a black box bolted onto your prompt. The connector page Asking & enforcing shows how to preview it in practice.
Where it fits
Injection is the “always on” half of governed memory: the decisions are there whether or not you think to ask. Ask is the “on demand” half, for when you want to interrogate the corpus directly.