ultimate-vibe-agent

An AI coding teammate for real product work.

Describe features in plain English.
Get context-aware code changes, diff review, verification loops, and safer team adoption.

---

Product Preview

Landing Experience

CLI Workflow

---

Why Teams Use This

• Build faster from product request to working code
• Reduce broken PRs with verification in the loop
• Keep control with manual diff approval before writes
• Work inside existing codebases, not toy examples
• Add governance via policy controls and audit logs

---

Core Product Capabilities

Natural Language Feature Delivery

You can ask:

"Add onboarding checklist with completion state and tests."

The agent plans and executes work against your project.

Real Codebase Awareness

It reads multiple files, tracks symbols, understands imports/usages, and maps dependencies.

Iterative Reliability Loop

It follows plan -> act -> verify and retries fixes when checks fail.

Human-Controlled Editing

Every write_file is shown as diff and needs approval.

Memory Across Sessions

Project conventions and architecture notes are persisted and reused.

Team Safety

Policy rules, secret detection, audit trails, and rollback support.

---

Architecture

flowchart LR
    U[User Prompt] --> A[Agent Runtime]
    A --> CI[Code Intelligence]
    A --> TO[Tool Runner]
    A --> ME[Memory Store]
    A --> PO[Policy Guardrails]
    TO --> FS[File System]
    TO --> SH[Shell Commands]
    TO --> VF[Verify Commands]
    CI --> IDX[Project Index]
    PO --> AU[Audit Log]
    VF --> ST[Stack Trace Parser]
    ST --> A
    A --> D[Diff Preview]
    D --> U

Loading

---

Quick Start

1) Install dependencies

npm install

2) Configure provider

Copy .env.example to .env and set:

GROQ_API_KEY=your_groq_api_key_here
VIBE_MODEL=moonshotai/kimi-k2-instruct-0905
VIBE_BASE_URL=https://api.groq.com/openai/v1

3) Run interactive mode

npm run dev

4) Run one-shot task

npm run dev -- --goal "add forgot-password flow with tests"

---

Usage Modes

CLI

npm install -g ultimate-vibe-agent
vibe-agent

In-chat commands:

• /task <goal>
• /config
• /help
• /exit

JavaScript SDK

import { CodingAgent, loadRuntimeConfig } from "ultimate-vibe-agent";

const config = loadRuntimeConfig(process.cwd(), {
  model: "moonshotai/kimi-k2-instruct-0905",
  baseUrl: "https://api.groq.com/openai/v1",
  maxIterations: 8,
});

const agent = new CodingAgent(config);

const ui = {
  info: console.log,
  warn: console.warn,
  error: console.error,
  ask: async () => "continue",
  confirm: async () => true,
};

await agent.runTask("refactor payment retries and add regression tests", ui);

Python SDK Wrapper

from vibe_agent_sdk import VibeAgentClient

client = VibeAgentClient(
    executable="vibe-agent",
    env={
        "GROQ_API_KEY": "your-key",
        "VIBE_MODEL": "moonshotai/kimi-k2-instruct-0905",
        "VIBE_BASE_URL": "https://api.groq.com/openai/v1",
    },
)

result = client.run_task(
    "optimize search endpoint and add test coverage",
    max_iterations=6,
    check=False,
)

print(result.returncode)
print(result.stdout)

Install the Python package:

pip install ultimate-vibe-agent

or

uv pip install ultimate-vibe-agent

---

Built-In Tooling

• list_files
• read_file
• grep
• run_command
• write_file (approval required)
• scan_project
• symbol_lookup
• find_references
• dependency_map
• memory_set
• memory_get

---

Safety and Governance

Default safeguards include:

• command policy checks
• blocked sensitive write paths (for example .env, .git/*)
• secret detection before write
• full session audit logs
• rollback option when unresolved failures remain

State artifacts:

• .vibe-agent/memory.json
• .vibe-agent/policy.json
• .vibe-agent/index/project-index.json
• .vibe-agent/audit/*.jsonl

---

Configuration

Variable	Purpose	Default
GROQ_API_KEY	Provider API key	empty
VIBE_API_KEY	Generic API key override	empty
VIBE_MODEL	Model name	moonshotai/kimi-k2-instruct-0905
VIBE_BASE_URL	OpenAI-compatible endpoint	https://api.groq.com/openai/v1
VIBE_MAX_ITERATIONS	Max loop rounds	6
VIBE_TOOL_TIMEOUT_MS	Tool timeout	120000
VIBE_MAX_TOOL_OUTPUT_CHARS	Output truncation limit	18000
VIBE_MAX_SCAN_FILES	Indexing limit	6000
VIBE_AUTO_REPAIR_ROUNDS	Retry cap for repeated failures	3
VIBE_AUTO_VERIFY	Auto-run verify commands	true
VIBE_STATE_DIR	Agent state directory	.vibe-agent

---

Build, Run, and Validate

npm run check
npm run build
npm run start

Python build:

npm run build:py

---

Landing Page

Serve locally:

npm run landing:serve

Open http://localhost:4173.

Vercel deployment:

• Root Directory: landing

---

Release

One-command release after version bump and commit:

release.cmd 0.1.1

Detailed checklist:

• docs/release-plan.md

---

License

MIT - see LICENSE.