The world's first governed coding agent.
No code ships without your approval. Ever.
Install · Why Viki Code · How It Works · Surfaces · Proof · Docs
Cursor. GitHub Copilot. Devin. Claude Code. They all operate on the same implicit promise: trust us. They generate code, they make changes, they touch your repositories — and they ask you to review after the fact, if at all.
Viki Code operates on an explicit guarantee: nothing happens without you.
It is the only autonomous coding system in existence where the governance layer — mandatory human approvals, worktree isolation, instant rollback, live policy validation, and a complete audit trail — is not a setting you toggle on. It is a foundation you cannot turn off.
╔═════════════════════════════════════════════════════════════════════════════╗
║ ║
║ The most powerful coding agent is not the fastest one. ║
║ It is the one your team can actually trust with your codebase. ║
║ ║
╚═════════════════════════════════════════════════════════════════════════════╝
# One command. Ready in under a minute.
git clone https://github.com/rebootix-research/viki-code.git
cd viki-code
python scripts/install.py --path .
vikiThat’s it. VIKI opens with a guided setup wizard, detects your provider, and drops you into a prompt-first session. No config files to hunt. No environment variables to memorize. Just type what you want done.
# Or install from the release wheel
pip install dist/viki_code-4.1.4-py3-none-any.whl
# Or run in Docker
docker pull ghcr.io/rebootix-research/viki-code:latest
docker run --rm -it ghcr.io/rebootix-research/viki-code:latest┌─────────────────────────────────────────────────────────────────────────┐
│ What happens when you use a normal coding agent: │
│ │
│ You → "Fix the auth bug" │
│ AI → [Makes 47 changes across 12 files] │
│ AI → "Done! Here's a summary." │
│ You → [Tries to understand what just happened to your codebase] │
│ You → [Finds three new bugs introduced during the "fix"] │
│ You → "How do I undo this?" │
│ AI → "..." │
└─────────────────────────────────────────────────────────────────────────┘
┌─────────────────────────────────────────────────────────────────────────┐
│ What happens when you use Viki Code: │
│ │
│ You → "Fix the auth bug" │
│ VIKI → [Plans the fix, scopes to affected files only] │
│ VIKI → [Shows you a clean diff before touching anything] │
│ VIKI → "Here's what I want to do. Approve?" ← YOU DECIDE │
│ You → "Yes" / "No" / "Change this part first" │
│ VIKI → [Executes exactly what was approved] │
│ VIKI → [Runs tests, validates, confirms green] │
│ VIKI → [Logs every action with a cryptographic record] │
│ You → [Can undo any of it in one command, any time] │
└─────────────────────────────────────────────────────────────────────────┘
The difference is not capability. The difference is control.
┌─────────────────────────────────────────────────────────────────────────────────┐
│ │
│ 1. You state intent │
│ ───────────────────────────────────────────────────────────────────────── │
│ viki> Fix the broken calculation in billing.py and make the tests pass │
│ │
│ │ │
│ ▼ │
│ 2. VIKI plans and scopes │
│ ───────────────────────────────────────────────────────────────────────── │
│ ┌─────────────────────────────────────────────────────────────────────┐ │
│ │ Planning... │ │
│ │ ├─ Analyzing repository structure │ │
│ │ ├─ Locating affected symbols: calculate_total(), apply_discount() │ │
│ │ ├─ Identifying impacted tests: test_billing.py, test_invoices.py │ │
│ │ └─ Scoping change to 2 files, 3 functions │ │
│ └─────────────────────────────────────────────────────────────────────┘ │
│ │
│ │ │
│ ▼ │
│ 3. VIKI shows you the diff before touching anything │
│ ───────────────────────────────────────────────────────────────────────── │
│ ┌─────────────────────────────────────────────────────────────────────┐ │
│ │ billing.py │ │
│ │ ───────────────────────────────────────────────────────────────── │ │
│ │ - total = price * quantity │ │
│ │ + total = round(price * quantity, 2) │ │
│ │ - discount = total / discount_rate │ │
│ │ + discount = total * (discount_rate / 100) │ │
│ └─────────────────────────────────────────────────────────────────────┘ │
│ │
│ │ │
│ ▼ │
│ 4. You approve (or reject, or modify) ◀── THIS IS MANDATORY │
│ ───────────────────────────────────────────────────────────────────────── │
│ ┌─────────────────────────────────────────────────────────────────────┐ │
│ │ 2 changes staged for billing.py │ │
│ │ Approve this change? [y/n/edit/explain] _ │ │
│ └─────────────────────────────────────────────────────────────────────┘ │
│ │
│ │ │
│ ▼ │
│ 5. VIKI executes, validates, confirms │
│ ───────────────────────────────────────────────────────────────────────── │
│ ┌─────────────────────────────────────────────────────────────────────┐ │
│ │ Executing approved changes... │ │
│ │ ✓ billing.py patched │ │
│ │ ✓ Running test_billing.py → 12/12 passed │ │
│ │ ✓ Running test_invoices.py → 8/8 passed │ │
│ │ ✓ Audit record written │ │
│ │ ✓ Rollback path preserved │ │
│ │ │ │
│ │ Task complete. Undo available with: viki rollback <session_id> │ │
│ └─────────────────────────────────────────────────────────────────────┘ │
│ │
└─────────────────────────────────────────────────────────────────────────────────┘
VIKI Code runs three specialized agents for every task. Not one model doing everything — three dedicated cognitive roles, each accountable for a distinct part of the pipeline:
┌───────────────────────────────────────────────────────────────────────────────┐
│ │
│ ┌─────────────────┐ ┌─────────────────┐ ┌─────────────────────────┐ │
│ │ │ │ │ │ │ │
│ │ PLANNER │───▶│ IMPLEMENTER │───▶│ VALIDATOR │ │
│ │ │ │ │ │ │ │
│ │ Understands │ │ Writes the │ │ Runs tests. Checks │ │
│ │ your repo. │ │ actual code │ │ policy. Confirms │ │
│ │ Scopes the │ │ changes. │ │ nothing is broken. │ │
│ │ work. Plans │ │ Respects the │ │ Refuses to sign off │ │
│ │ the safe path. │ │ plan. Stays │ │ if validation fails. │ │
│ │ │ │ in scope. │ │ │ │
│ └─────────────────┘ └─────────────────┘ └─────────────────────────┘ │
│ │
│ Repo intelligence at every stage: │
│ Symbol lookup · Import graph · Impact analysis · Test targeting │
│ │
└───────────────────────────────────────────────────────────────────────────────┘
VIKI Code does not stuff your entire codebase into a context window and hope. It builds a live model of your repository:
# See what VIKI knows about your repo
viki repo "auth migration" --path .
# Find symbols across the codebase
viki symbols "normalize_account" --path .
# Understand what changes to a file would affect
viki impact --changed-file viki/api/server.py --path .┌─────────────────────────────────────────────────────────────────────────┐
│ viki impact --changed-file src/auth/session.py │
├─────────────────────────────────────────────────────────────────────────┤
│ │
│ Direct dependents (3): │
│ src/api/routes/login.py ← imports session.validate() │
│ src/middleware/auth.py ← imports session.refresh() │
│ tests/test_session.py ← 14 test cases affected │
│ │
│ Indirect dependents (7): │
│ src/api/routes/dashboard.py │
│ src/api/routes/settings.py │
│ ... │
│ │
│ Recommended test scope: │
│ pytest tests/test_session.py tests/test_auth.py │
│ │
└─────────────────────────────────────────────────────────────────────────┘
# Bug fixes
viki run "Fix the broken calculation in billing.py and make tests pass" --path .
# Refactors
viki run "Refactor auth naming consistently across the codebase and keep all tests green" --path .
# Migrations
viki run "Migrate the old consumer to the new API and run the relevant tests" --path .
# Repo intelligence
viki run "Inspect this repo and give me a full summary of the architecture" --path .
# Multi-step work
viki run "Add input validation to all public API endpoints and write tests for each" --path .# See what's running or completed
viki status . --session-id <session_id>
# Review the diff from any session
viki diff <session_id> --path . --rendered
# Undo everything from a session
viki rollback <session_id> --path .
# Export a patch bundle
viki patch <session_id> --path .┌──────────────────────────────────────────────────────────────────────────────┐
│ │
│ ┌───────────┐ ┌───────────┐ ┌───────────┐ ┌───────────┐ │
│ │ │ │ │ │ │ │ │ │
│ │ CLI │ │ API │ │ VS Code │ │ Messaging │ │
│ │ │ │ │ │ │ │ │ │
│ └───────────┘ └───────────┘ └───────────┘ └───────────┘ │
│ │ │ │ │ │
│ Direct terminal Local HTTP Repo-aware Telegram & │
│ operator use. REST API for extension. WhatsApp. │
│ Full session orchestration Task submission Approve, │
│ control. and CI/CD. and approval reject, diff │
│ integration. from your IDE. from your phone. │
│ │
│ └─────────────────────────────┘ │
│ Same execution core. │
│ Same governance model. │
│ All surfaces. │
│ │
└──────────────────────────────────────────────────────────────────────────────┘
viki # Interactive prompt-first session
viki run "task" --path . # Direct task execution
viki setup # Provider setup wizard
viki setup --repair # Repair broken setup
viki providers # List configured providers
viki doctor . # Health check your workspace# Start the local API server
viki up . --host 0.0.0.0 --port 8787GET /healthz Health check
GET /protocol API version and capabilities
POST /runs Submit a new task
GET /runs/{id} Task status
GET /runs/{id}/diff Diff for a task
GET /runs/{id}/result Final result
GET /approvals Pending approvals
POST /approvals/{id} Approve or reject
GET /repo/symbols?q=... Symbol lookup
GET /repo/impact?path=... Impact analysis
viki ide vscode . # Generate workspace integration
viki ide vscode-extension . # Full extension setup/status <session_id> See what's running
/approvals List pending approvals
/approve <id> Approve a staged change
/reject <id> Reject a staged change
/diff <session_id> See the diff on your phone
/patch <session_id> Download the patch bundle
/rollback <session_id> Undo everything
VIKI Code works with every major provider. Switch in one command. No code changes.
┌────────────────────────────────────────────────────────────────────────┐
│ Supported Providers │
├────────────────────────────────────────────────────────────────────────┤
│ │
│ ✓ Anthropic (Claude Sonnet, Claude Opus) │
│ ✓ OpenAI (GPT-4o, o1, o3) │
│ ✓ NVIDIA NIM (Kimi K2.5, GLM-5, Llama 3.3) │
│ ✓ DashScope / Qwen (Qwen3-Coder, Qwen3) │
│ ✓ OpenRouter (Any model via unified API) │
│ ✓ Azure OpenAI (Enterprise deployments) │
│ ✓ Ollama (Local models, fully offline) │
│ ✓ Generic OpenAI-compatible endpoints │
│ │
│ First-class NVIDIA preset: choose NVIDIA → pick Kimi K2.5 → │
│ paste key → start prompting. That's it. │
│ │
└────────────────────────────────────────────────────────────────────────┘
Viki Code is designed to be useful under autonomy without pretending autonomy should be ungoverned.
┌─────────────────────────────────────────────────────────────────────────────────┐
│ GOVERNANCE BY DEFAULT. NOT BY CONFIGURATION. │
├─────────────────────────────────────────────────────────────────────────────────┤
│ │
│ ┌──────────────────────┐ Every task runs in an isolated worktree. │
│ │ Worktree Isolation │ The repository is never touched until you │
│ └──────────────────────┘ approve. Never. │
│ │
│ ┌──────────────────────┐ Approval-aware action flow for every │
│ │ Approval Gate │ consequential operation. The gate is │
│ └──────────────────────┘ architectural. Not a setting. Not a flag. │
│ │
│ ┌──────────────────────┐ Diff preview and patch bundle before │
│ │ Diff Preview │ anything is applied. See exactly what │
│ └──────────────────────┘ will change before it changes. │
│ │
│ ┌──────────────────────┐ One command. Any session. Complete │
│ │ Instant Rollback │ reversion. No data loss. No partial │
│ └──────────────────────┘ states. Clean every time. │
│ │
│ ┌──────────────────────┐ Every action logged with attribution. │
│ │ Audit Trail │ Cryptographic record of everything │
│ └──────────────────────┘ VIKI touched. Verifiable on demand. │
│ │
│ ┌──────────────────────┐ Sensitive paths, keys, and outputs │
│ │ Redacted Logs │ automatically scrubbed from all │
│ └──────────────────────┘ proof artifacts and session logs. │
│ │
└─────────────────────────────────────────────────────────────────────────────────┘
We publish everything. Every benchmark. Every live run result. Every test. No black box claims.
┌────────────────────────────────────────────────────┬──────────┬────────────┐
│ Validation Signal │ Result │ Version │
├────────────────────────────────────────────────────┼──────────┼────────────┤
│ Local regression suite │ 76/76 │ 4.1.4 │
│ Live validation suite (fresh repos) │ 9/9 │ 4.1.4 │
│ Generic CLI live wins │ 7/7 │ 4.1.4 │
│ Public live benchmark slice │ 8/8 │ 4.1.4 │
│ Public offline benchmark slice │ 8/8 │ 4.1.4 │
│ Human-style install validation │ Passed │ 4.1.4 │
│ WSL-isolated live execution │ Passed │ 4.1.4 │
└────────────────────────────────────────────────────┴──────────┴────────────┘
Honest limitation: Viki Code currently trails some baselines on time-to-green even when it completes the task successfully. We document this openly. Speed is a known work item and is being addressed in the active development cycle.
→ Full Proof Report · Benchmark Results · Live Run Results
┌─────────────────────────────────────────────────────────────────────────────────┐
│ │
│ Platform & Infrastructure Teams │
│ ────────────────────────────────────────────────────────────────────────── │
│ You need AI execution that doesn't become a liability. VIKI Code gives │
│ you automation with approval gates, audit trails, and rollback built in. │
│ │
│ Engineering Teams in Large Repos or Monorepos │
│ ────────────────────────────────────────────────────────────────────────── │
│ VIKI Code's repo intelligence indexes your structure, understands │
│ your symbol graph, and scopes work correctly — instead of hallucinating │
│ changes across files it never read. │
│ │
│ Teams That Need Approvals, Rollback, and Proof │
│ ────────────────────────────────────────────────────────────────────────── │
│ Compliance requirements. Security reviews. Regulated industries. │
│ Viki Code was built for exactly this. Governance is not an add-on. │
│ It is the entire design philosophy. │
│ │
│ Builders Who Want a Serious Local Agent, Not a Hosted Chat UX │
│ ────────────────────────────────────────────────────────────────────────── │
│ Your data stays on your machine. Your code stays in your perimeter. │
│ Your API key goes to a provider of your choice. Nothing else. │
│ │
└─────────────────────────────────────────────────────────────────────────────────┘
Cursor Copilot Devin Claude Code VIKI CODE
────── ─────── ───── ─────────── ─────────
Mandatory approvals ✗ ✗ ✗ ✗ ✓
Instant rollback ✗ ✗ ✗ ✗ ✓
Cryptographic audit ✗ ✗ ✗ ✗ ✓
Worktree isolation ✗ ✗ ⚠ ✗ ✓
Diff preview gate ⚠ ✗ ⚠ ✗ ✓
Local-first ✗ ✗ ✗ ✗ ✓
Multi-agent runtime ✗ ✗ ✓ ✓ ✓
Repo intelligence ✓ ✓ ✓ ✓ ✓
Messaging surfaces ✗ ✗ ✗ ✗ ✓
Provider agnostic ✗ ✗ ✗ ✗ ✓
Open source ✗ ✗ ✗ ✗ ✓
VIKI ships with a premium terminal presentation layer. In a capable terminal it renders:
╔══════════════════════════════════════════════════════════════════════════════╗
║ ║
║ VIKI Code v4.1.4 rebootix-research/viki-code ║
║ Provider: Anthropic · Model: claude-sonnet-4-20250514 ║
║ Workspace: /home/user/projects/my-api Branch: main ║
║ ║
╠══════════════════════════════════════════════════════════════════════════════╣
║ ║
║ Session: a3f7d291 · Started: 14:32:01 · Status: Active ║
║ ║
║ ┌──────────────────────────────────────────────────────────────────────┐ ║
║ │ ▶ Planner Scoping fix to billing.py — 2 functions affected │ ║
║ │ ▶ Implementer Writing patch... │ ║
║ │ ◷ Validator Awaiting implementer │ ║
║ └──────────────────────────────────────────────────────────────────────┘ ║
║ ║
╚══════════════════════════════════════════════════════════════════════════════╝
viki> _
viki --theme premium run "task" --path . # Full premium render
viki --theme contrast run "task" --path . # High contrast
viki --plain run "task" --path . # Plain output for CI
viki --force-rich --theme premium doctor . # Force themed outputviki/ Core runtime, repo intelligence, orchestration,
│ API, IDE tooling, integrations
│
├── agents/ Planner, Implementer, Validator agents
├── repo/ Repo intelligence — symbols, impact, indexing
├── api/ Local HTTP REST API server
├── ide/ VS Code extension generation
├── integrations/ Telegram, WhatsApp webhook handlers
├── governance/ Approval gate, audit trail, rollback engine
└── ui/ Premium terminal presentation layer
scripts/ Install, validation, live-run, release helpers
tests/ Unit, integration, CLI, API, regression
BENCHMARK_RESULTS/ Machine-readable benchmark artifacts
LIVE_RUN_RESULTS/ Live validation artifacts
docs/ Benchmark board and documentation
| Document | Description |
|---|---|
| <PROOF_REPORT.md> | Full validation and benchmark evidence |
| <RELEASE_NOTES.md> | What changed in each release |
| <AGENTS.md> | How the multi-agent runtime works |
| <CONTRIBUTING.md> | How to contribute to Viki Code |
| <SECURITY.md> | Security policy and disclosure |
| <BENCHMARK_RESULTS/> | All benchmark artifacts |
| <LIVE_RUN_RESULTS/> | All live run artifacts |
Viki Code is open source and welcomes contributions. Read <CONTRIBUTING.md> before opening a pull request.
If you find a security issue, read <SECURITY.md> and disclose responsibly.
If Viki Code is relevant to your stack — star the repo. It helps more than you think.
╔═══════════════════════════════════════════════════════════════════════╗
║ ║
║ The most powerful coding agent in the world ║
║ is not the fastest one. ║
║ ║
║ It is the one you can actually trust ║
║ with your most important repositories. ║
║ ║
╚═══════════════════════════════════════════════════════════════════════╝
Built by Rebootix Artificial Intelligence Research and Development
Sovereign AI Infrastructure · United Arab Emirates · 2026