Local scan
npx ship-safe audit . --no-aiStays on your machine
The 29 built-in agents, dependency checks, scoring, policy evaluation, and report generation run locally. No AI provider is called.
Know where your code goes.
Ship Safe CLI is local-first. Remote AI analysis and hosted workflows are separate modes with explicit data boundaries and controls.
Verified against Ship Safe CLI v9.5.2 · July 19, 2026
29built-in agents run locally
0API keys required for core scans
1 flag
--no-ai forces offline scanningReal product output
Findings remain reviewable.
Local scans can produce structured findings and reports. Hosted review is a separate workflow you choose when you need history, investigation, and team collaboration.
Processing modes
Local by default. Remote when selected.
Having a supported provider key in your environment may enable optional AI classification unless you pass --no-ai.
Provider analysis
npx ship-safe audit . --provider <name>Your selected provider
Ship Safe sends finding metadata and bounded matched-code context for classification or deep analysis. Common credential patterns are masked before provider-bound prompts.
GPT-Red
npx ship-safe red-team . --gpt-red --provider <name>Your selected provider
The red-team harness sends bounded agent-readable files and relevant configuration so attacker, defender, and judge scenarios can reason across real repository context.
Hosted workflows
Dashboard, reports, and PR GuardianShip Safe cloud services
Repository metadata, scan results, findings, and reports are processed as needed to provide the hosted feature you explicitly use.
What remote AI modes receive
- Standard classification includes finding metadata and a matched excerpt capped at 100 characters.
- Deep analysis can include a bounded code window around a finding to reason about reachability and sanitization.
- Standard GPT-Red selects up to 12 relevant files, 6,000 characters per file, and 60,000 characters total.
- Kimi K3 long-context mode raises those caps to 60 files, 12,000 characters per file, and 250,000 characters total.
Masking before provider calls
Provider-bound classification, deep-analysis, and GPT-Red context applies best-effort masking for private keys, authorization headers, common secret assignments, and recognizable provider-token formats.
Masking is defense in depth, not a guarantee that arbitrary sensitive business data can be identified. Use --no-ai when repository content must not leave the machine.
Your key, your endpoint
The CLI calls the provider configured through your environment or command options. Provider API keys authenticate directly to that endpoint; they are not uploaded to the Ship Safe dashboard by the local CLI.
For local model inference, use a supported localhost provider such as Ollama or LM Studio.
What is stored on disk
Incremental scan context, score history, accepted baselines, fix logs, and time-limited LLM response caches can be written under .ship-safe/ in the scanned project.
Add that directory to .gitignore when you do not want local scan state committed.
Strict local mode
Keep the scan entirely offline.
Use the explicit control when source code or findings cannot be sent to any external model.
$ npx ship-safe audit . --no-aiCommon questions
Security, without ambiguity.
Does Ship Safe CLI send source code to the cloud?
The 29 built-in agents can run locally without sending source code to Ship Safe or an AI provider. Provider-backed analysis and GPT-Red send bounded repository context to the provider you select. Use --no-ai to force offline scanning.
Does Ship Safe CLI require an API key?
No API key is required for local scanning. A provider API key is required only when you explicitly use remote AI classification, deep analysis, or provider-backed GPT-Red workflows.
How do I run Ship Safe completely offline?
Run npx ship-safe audit . --no-ai. The local agent pool, dependency checks, scoring, policies, and report generation remain on your machine.