Bounded LoopsIndependent Gates for Bounded Agent Loops
Run a worker under declared limits and require an independent acceptance gate before completion. Start with the shipped keyless test example, then inspect its persisted ledger and the research paper.
A claim of done needs a separate check
A worker can claim success while its output still fails a test. A bounded loop gives the acceptance check its own authority, records the gate verdict and stops at the declared limits.
Key Capabilities
Independent acceptance
The shipped demo uses a real pytest gate with a stub worker; the worker's claim does not decide completion.
Declared bounds
Limit attempts, runtime and other budgets before running. Inspect the exact contract and supported runner behavior.
Persisted receipts
Retain the run metadata and a separate head digest, then verify the ledger chain, anchor and completeness.
Graphs and repair
Compose gated nodes with repair routes. Read the graph contract and paper assumptions before adopting a workflow.
The 1 October 2026 local demo showed an ungated false GREEN claim, then a gated run with six pytest passes and an anchored complete receipt. It is not a hosted-agent accuracy, speed or savings benchmark. Research results remain scoped to the paper's protocol.
Get Started
SuperLocalMemory
Local-first Memory for AI Agents
Store and recall project context across AI agent sessions through MCP or CLI, with workspace isolation, explicit operating modes and auditable retrieval.
AgentAssert
Runtime Behavioral Contracts for AI Agents
Define YAML rules and check supplied structured agent state in Python. Place checks at an action boundary and raise on hard violations.