TOOLS
Open infrastructure for controllable, reproducible AI-agent execution.
Ghost and Replay are two independent open-source research tools around autonomous-agent infrastructure: Ghost constrains what an agent can reach and can expose controlled synthetic resources; Replay records execution so technical state can be inspected, restored, branched and compared.
TWO DIFFERENT QUESTIONS
Control the environment. Preserve the evidence.
Ghost — What may the agent access — and what happens when it touches a decoy?
Replay — What happened during the run — and can the technical state be reconstructed?
Ghost and Replay are complementary research directions, not presented here as a single integrated product. Their current implementations remain separate repositories with separate guarantees.
ALLOW · DENY · SHADOW
rappidAI Ghost
A deception-aware security runtime with deterministic ALLOW, DENY and SHADOW policy outcomes.
CURRENT IMPLEMENTATION
- — Docker-isolated sessions
- — Synthetic agent home and decoy resources
- — Controlled HTTP/HTTPS egress
- — Evidence-based decoy-access events
- — Dynamic network containment
RECORD · RESTORE · BRANCH · DIFF
rappidAI Replay
Local-first infrastructure for recording, restoring, branching, rerunning and diffing AI-agent executions.
CURRENT IMPLEMENTATION
- — Generic execution recorder
- — Verified restore
- — Branch and controlled rerun
- — Deterministic multi-dimensional diff
- — SQLite metadata plus content-addressed state
EVIDENCE BOUNDARY
Claims are tied to reviewed source revisions.
The pages below distinguish implemented behavior from architecture targets and link technical claims to pinned repository revisions and architecture records.