Ideas and Memory
Ideas are the durable knowledge objects inside an aeqi company. Memory is what the company becomes as ideas, sessions, events, quests, and outcomes accumulate.
This is where work compounds.
What counts as an idea
An idea can be:
- a strategy
- a fact
- a procedure
- a customer note
- a file
- a decision
- a policy
- a research finding
- a product requirement
- an agent identity note
- a standing instruction
- a reusable operating pattern
The name is intentionally broad. The company needs one universal noun for context.
Why memory matters
Most companies leak context. Decisions vanish into chat, procedures become stale, and new contributors start from zero.
aeqi treats memory as operating infrastructure. Agents can retrieve what the company knows, add what they learn, and carry context forward into future work.
That makes the company smarter over time.
How ideas activate
There is no "mode" flag on an idea. Activation is event-driven. When a session
starts, a session:start handler in the runtime assembles the relevant ideas —
an agent's identity and standing instructions (ideas tagged identity +
evergreen) plus anything retrieved for the work at hand — and appends them to
the system prompt. "Always-loaded" is just a convention layered through that
event, not a property of the row.
So agent identity reads as evergreen identity ideas; company strategy as evergreen ideas; quest-specific context as ideas attached to the quest. The mechanism is the same retrieval-and-assembly path in every case. For the developer view of scopes, ranking, and the MCP/REST surface, see Ideas.
Memory and accountability
Memory is not only for better answers. It is the beginning of operating truth.
When a company preserves:
- the instruction
- the context
- the role
- the agent
- the work trace
- the outcome
then contribution becomes inspectable. That is the foundation for future compensation, governance, authority, and capital allocation.