FAQ
Is SMALL a framework?
No. SMALL is a protocol. Tools can implement it, but the contract is the product.
Is SMALL an agent?
No. SMALL makes agent work resumable and auditable. Execution is handled by whatever tool you already use.
Do I need a server or database?
No. SMALL is file-based. Git is enough.
Do I have to use the small CLI?
No. The CLI is a reference implementation. Any tool can produce valid SMALL artifacts.
Does SMALL prevent failure?
No. SMALL assumes failure. It prevents failure from destroying context and continuity.
Can I use SMALL with Cursor / Claude / ChatGPT?
Yes. SMALL is designed to sit underneath any model or editor.
Is SMALL single-agent?
V1 is single-writer. Mutating CLI commands in one checkout are protected by a local cooperative lock; disconnected clones still require external coordination and explicit Git merging.
V2 remains solo by default, but it can opt into collaborative mode. Each writer uses an attributed session and immutable event paths. A deterministic reducer detects semantic conflicts and blocks strict validation until a complete, expected-frontier resolution is recorded.
Does SMALL support multi-agent workflows?
V1 does not model concurrent writers. V2 can record and reconcile multiple sessions inside one project after an explicit collaborative mode transition.
Execution orchestration remains external. Agents can be scheduled through:
- Human-defined workflows
- CI pipelines
- Message queues
- Or higher-level orchestration systems
SMALL records causal state and refuses unresolved ambiguity. It does not run agents, assign tasks, provide distributed leases, or merge source code.
Can I use Claude Code, Cursor, or other AI tools with SMALL?
Yes.
AI tools may be used as decision assistants, but they do not execute actions directly.
In SMALL:
- The AI proposes actions
- The human or system approves intent
- The SMALL CLI executes explicitly defined commands
- All execution is recorded in artifacts
SMALL governs execution. AI assists reasoning.