Memory that compounds: how Spawn agents learn your business
Every session makes the next one better. How persistent memory turns Spawn from a tool into a teammate that knows your context.
The first time you use Spawn, it's fast but generic. It doesn't know your team's naming conventions, which vendors you work with, what your commercial terms look like, or that you always round invoice totals to two decimals for accounting but four for FX reporting.
By the tenth time, it should know all of that. That's the difference between a tool and a teammate — and it's what persistent memory is for.
How memory works in Spawn
After every session, a small extraction agent reads the conversation and decides what's worth remembering. Not everything survives — trivial facts, one-off requests, and ephemeral details get dropped. What stays are durable patterns: preferred formats, recurring entities, your working style, the playbooks you reach for.
Those facts land in a typed memory store, indexed by topic and scoped to the right boundary (you, your team, your workspace). The next time a relevant agent runs, the memories in scope get injected into its system prompt before it starts working.
Three kinds of memory
- Facts — "Our standard NDA term is 3 years mutual." "Team email for procurement is ops-procurement@acme.com." Recalled verbatim.
- Preferences — "Always produce decks in 16:9 with our brand palette." "Default report format is markdown, not PDF." Applied automatically.
- Playbooks — multi-step routines the agent has run before successfully and will reuse next time. Named, versioned, editable.
“The first time you use Spawn, it's fast but generic. By the tenth time, it should know your team.
Forking and sharing
Memory lives at three scopes: individual, team, and workspace. When you add a teammate, their agents inherit the workspace memory automatically. When you fork an agent to a new project, you choose which memories carry over — keep the commercial preferences, drop the project-specific facts.
We think of it the way people think of institutional knowledge: some things are yours, some things are the team's, some things are the company's.
Why this compounds
A traditional SaaS product hits a ceiling at onboarding. Once your team is trained and the templates are configured, there's no more leverage. An agent with memory has the opposite dynamic — every session makes the next one better. The longer you use Spawn, the less setup each new task requires, and the more the agent handles without prompting.
This is the quiet lever behind agent-first software. Configuration used to be a one-time cost. Now it's a continuous return.