Your agents. Your environment. Your evidence.

Two open-source tools for agent infrastructure. Ghost controls access to the environment. Replay records execution and preserves inspectable technical state.

rappidAI Ghost: v0.3.0 · Integrated agent security · Experimental
rappidAI Replay: Experimental · Open source
Source: GitHub · rappidAI-Research

Two different questions

Control the environment. Preserve the evidence.

Ghost and Replay are complementary research directions, not presented here as a single integrated product. Their current implementations remain separate repositories with separate guarantees.

	rappidAI Ghost	rappidAI Replay
Question	What may the agent access — and how should suspicious input, sensitive actions and runtime risk be handled?	What happened during the run — and can the technical state be reconstructed?
Scope	An integrated security runtime for autonomous agents. Isolation, SHADOW deception, prompt-injection signals, scoped ASK decisions, runtime limits and evidence work together behind the same ghost run flow.	Local-first infrastructure for recording, restoring, branching, rerunning and diffing AI-agent executions.
Core actions	ALLOW · DENY · SHADOW · ASK	RECORD · RESTORE · BRANCH · DIFF

ghostv0.3.0

$ mkdir -p ../ghost-demo$ cd ../ghost-demo$ ghost init$ ghost run -- echo "hello from ghost"

ALLOW · DENY · SHADOW · ASK

rappidAI Ghost

An integrated security runtime for autonomous agents. Isolation, SHADOW deception, prompt-injection signals, scoped ASK decisions, runtime limits and evidence work together behind the same ghost run flow.

v0.3.0 · Integrated agent security · Experimental

CURRENT IMPLEMENTATION

Bounded Prompt-Injection Guard and deterministic trust context
ALLOW / DENY / SHADOW / ASK with exact session-scoped approval
Non-root Docker isolation plus memory, CPU, PID, tmpfs and runtime limits
Exact-hostname HTTP/HTTPS egress with containment taking precedence over approval
Evidence-based summaries, provenance, incidents and crash recovery
25-scenario adversarial GhostBench release gate: 25 PASS / 0 FAIL / 0 SKIP

Explore tool Inspect the source v0.3.0 Release

rappid replayCLI

$ rappid replay record -- <command>$ rappid replay verify <state-id>$ rappid replay restore <state-id>$ rappid replay branch <state-id>$ rappid replay diff <left-session-id> <right-session-id>

RECORD · RESTORE · BRANCH · DIFF

rappidAI Replay

Local-first infrastructure for recording, restoring, branching, rerunning and diffing AI-agent executions.

Experimental · Open source

CURRENT IMPLEMENTATION

Generic execution recorder
Verified restore
Branch and controlled rerun
Deterministic multi-dimensional diff
SQLite metadata plus content-addressed state

Explore tool Inspect the source

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.