Skip to content

Repository files navigation

Kryptic

CITestsBenchmarkNodeLatest releaseLast commitStarsForksOpen issuesLanguage

Kryptic v1 is a local-first AI engineering agent CLI. It prioritizes a reliable workspace boundary, focused repository inspection, provider separation, direct command execution, bounded repair, verified skills, and evidence-based tests before any hosted interface is added. It runs on your own computer, uses your own provider keys, and does not require a hosted Kryptic account.

Quick start

The fastest path for a fresh checkout is:

git clone https://github.com/adnqcr7-code/kryptic.git
cd kryptic
npm run v1:test
node src/cli.js start

The first interactive start asks for a provider, stores the key outside the workspace, checks local prerequisites, and opens chat. Run the complete local v1 release check with one command:

npm run v1:test

For everyday use, the concise commands are:

node src/cli.js chat
node src/cli.js setup
node src/cli.js "inspect the repository and improve the tests"
node src/cli.js fix "repair the failing test and verify it"
node src/cli.js test
node src/cli.js doctor
node src/cli.js history
npm run demo

See DEMO.md for the five-minute walkthrough and offline demo. See ROADMAP.md for measurable next milestones. See V1-AGENT-TEST.md for the agent-test contract and acceptance criteria. See CONTRIBUTING.md for safe contribution expectations and SECURITY.md for the prototype’s threat-model boundaries.

Current status

The project is an early local prototype. It currently provides:

CapabilityStatus
Local CLI helpAvailable
Provider status for Google, OpenAI, and ClaudeAvailable
Workspace inspectionAvailable
Focused repository searchAvailable
Recursive repository context inspectionAvailable
Path traversal protectionAvailable
Secret-looking file protectionAvailable
Guarded command policyAvailable
Provider-backed planningAvailable when a provider key is configured
Bounded structured run commandAvailable when a provider key is configured
Exact-match patching and approval-gated editsAvailable
Run-history audit recordsAvailable
Provenance-aware local memoryAvailable
Offline demo without an API keyAvailable
Deterministic safety benchmarkAvailable
Interactive approvalsAvailable in TTY mode
Resumable run statuses and audit eventsAvailable
Bounded plan–act–verify–repair loopAvailable foundation
Transactional rollback and diff previewsAvailable
Interrupted-run resumeAvailable via resume
Strict response-size and criteria limitsAvailable
Linked verify_workspace agent toolAvailable
Structured verification evidence and failure classificationAvailable
Autonomous long-horizon coding loopIn progress

Requirements

Node.js 20 or newer is required. No npm dependencies are currently needed.

Usage

node src/cli.js --help
node src/cli.js providers
node src/cli.js inspect .
node src/cli.js search "TODO" --workspace .
node src/cli.js skills --workspace .
node src/cli.js skills "test repair" --workspace .
node src/cli.js ask "Explain this repository and plan a small test improvement" --provider openai --workspace .
node src/cli.js run "Inspect the repository and propose a safe test improvement" --provider openai --workspace .
node src/cli.js run "Fix the failing test and verify the repair" --provider openai --workspace . --approve --repair
node src/cli.js benchmark
node src/cli.js resume <run-id> plan.json --interactive --workspace .
npm test

Provider credentials are read from environment variables and are never written by Kryptic into the repository. The supported variables are GEMINI_API_KEY, OPENAI_API_KEY, and ANTHROPIC_API_KEY. Model overrides are KRYPTIC_GOOGLE_MODEL, KRYPTIC_OPENAI_MODEL, and KRYPTIC_CLAUDE_MODEL.

Safety defaults

Kryptic resolves paths against the selected workspace and rejects traversal outside it. It refuses secret-looking files such as .env, private-key files, and package-manager credential files. The command policy blocks destructive file operations, network-download commands, shell escapes, risky git operations, and unapproved commands. The policy is intentionally conservative and will expand only when tests justify the change.

Planned next milestone

The --repair run mode now connects provider-generated replacement plans to the bounded repair loop while preserving schema validation, approvals, rollback, and evidence-based success criteria. Verified procedural skills can be promoted only from status: verified evidence and discovered with skills. The model must never be allowed to invent a successful test result; verification comes from the local executor.

Get involved

The most useful early contribution is a reproducible issue or a small fixture that reveals a safety, repair, provider, browser, or cross-platform failure. Please use the bug template or feature template. Testers can use the feedback form to report onboarding friction and structured evidence. Do not coordinate votes or post API keys.

Source material

The older Kairos repository is kept in a separate reference directory during development. Kryptic will reuse sound local-first and safety patterns without copying the entire Kairos control plane.

Website

The Kryptic website is hosted separately from this repository: https://kryptic-site-meme-56ed.vercel.app

About

Local-first AI engineering agent with bounded execution, verification, repair, browser control, and safety-first workflows

Topics

Resources

Contributing

Security policy

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages