Knowledge & memory
An agent that forgets everything between turns is exhausting to work with. withvibe gives agents durable memory at four scopes, each written by the agent itself and read back automatically on every turn.
The four layers
Each layer answers a different question — "how should I behave?", "what do I know about this person?", "what's true about this env?", "what's true for the whole team?" — and each has its own scope:
- Agent skills — reusable rules and recipes that change how one agent behaves. Scoped to a single agent, either workspace-wide or to one env. See self-learning skills.
- Member memory— private notes about a specific person (their preferences, decisions they've made). Only surfaces when that person is the one talking.
- Env knowledge — facts true for one environment: a quirk in its compose setup, which port a service really listens on, a gotcha in its seed data. Shared by everyone in that env.
- Workspace knowledge — ground-truth facts for the whole team, true across every env: conventions, where things live, how your product works.
How it gets written
The agent writes memory itself, with a dedicated tool per layer, when it learns something worth keeping. It routes each fact to the right tier — a rule about its own behavior becomes a skill, a preference about you becomes member memory, an env-specific gotcha becomes env knowledge, a team-wide fact becomes workspace knowledge. You don't have to manage this; it happens as a side effect of working.
One important shortcut: when you answer a questionan agent asked you, your answer is saved as member memory automatically — so the agent reads it next turn and doesn't ask again.
How it gets read
Memory is injected into the agent's context automatically at the start of every turn — there's no "load memory" step the agent has to remember to do. Member memory is added only for the current speaker; env and workspace knowledge are always present; an agent's skills come along wherever that agent runs. A read-only mirror of memory is also written into the env so the agent can grep it like any other file.
What persists, and where
- Member memory and workspace knowledge are workspace-scoped — they follow you and the team across every env.
- Env knowledge and env-scoped skills are bound to one env; workspace-scoped skills travel with their agent.
- Conversation history persists per session, handled by the underlying Claude session rather than flattened into the prompt.