AI Lab — ML/LLM Engineering Platform

One platform. Ten providers. Zero friction. The missing engineering layer for AI developers who live in OpenCode.

Live Demo → · Docs → · Philosophy → · Blueprint → · Support →

---

AI Lab is a unified, local-first ML/LLM Engineering platform — part observability, part gateway, part model catalog, and part cost analytics — designed from day one for developers who use OpenCode as their AI coding companion. Runs on your machine (planned: Tauri desktop app). Manage 10+ providers, discover 100+ models, track experiments, sync config in one command, and never context-switch between tools again.

---

Why AI Lab?

Building production AI applications forces teams to juggle 3–5 separate tools: observability (Helicone), experiments (W&B), model playgrounds (Langfuse), cost tracking, and config management. Data fragments across silos, costs pile up on per-seat pricing, and developers waste time stitching workflows together.

AI Lab solves this with one premise: the platform should adapt to your workflow, not the other way around.

Problem	AI Lab Solution
Fragmented observability	Unified dashboard — tokens, cost, latency, traces in one view
Model vendor lock-in	Registry for 10+ providers, switch models in seconds
Cost overruns	Real-time tracking + projections + budget alerts
Tool sprawl	One local-first instance replaces 5+ tools
Config sync overhead	config sync opencode bridges your AI editor in one command
Per-seat licensing	Free. Unlimited users. Self-hosted.

---

Features

Area	Capabilities	Status
Provider Management	10 built-in providers, multi-instance per provider, connection testing, model scanning	✅
Model Catalog	Search, compare side-by-side, recommend by task/budget, aliases	✅
Config Sync	Sync to OpenCode, Cursor, Continue, Aider — auto-generated provider config	✅
Experiment Tracking	Create, run, save results, compare across models and prompts	✅
Cost Analytics	Usage breakdown by provider/model, projections, budget alerts	✅
Orchestration	OMO agent detection, skill viewer, OpenCode config reader	✅
Export/Import	Full platform config as JSON — share with your team	✅
Env Detection	Auto-import from PROVIDER_*_API_KEY environment variables	✅

---

OpenCode-Native

AI Lab is the only ML/LLM engineering platform that treats OpenCode as a first-class citizen.

┌─────────────────────────────────────────────────────────────┐
│                    AI Lab + OpenCode                        │
├─────────────────────────────────────────────────────────────┤
│                                                             │
│  ml-engine config sync opencode                             │
│       ↓                                                     │
│  Reads provider instances from SQLite store                 │
│       ↓                                                     │
│  Generates ~/.config/opencode/opencode.json                 │
│  Generates ~/.local/share/opencode/auth.json                │
│       ↓                                                     │
│  OpenCode instantly has access to all your providers        │
│  and models — no manual config editing                      │
│                                                             │
└─────────────────────────────────────────────────────────────┘

What makes it native:

• One-command config sync — ml-engine config sync opencode writes provider config and auth keys directly into OpenCode's native JSON format. No third-party scripts, no manual editing.
• OMO (Oh My OpenCode) aware — AI Lab reads agents from ~/.config/opencode/agents/ and skills from ~/.config/opencode/skills/, giving you a dashboard of your entire OpenCode orchestration setup.
• Forked AI SDK — @ai-lab/openai-compatible (forked from @ai-sdk/openai-compatible) includes the minChunkSize fix that stops SSE chunk fragmentation — the same fix that powers smooth streaming in OpenCode itself.
• Works the way you work — add a provider in the CLI or Web UI, sync to OpenCode, and your editor is ready. No file hunting, no config format research.

---

Architecture

┌─────────────────────────────────────────────────────────────────┐
│                         AI Lab Platform                         │
├─────────────────────────────────────────────────────────────────┤
│                                                                 │
│  ┌─────────────────────┐     ┌──────────────────────────────┐  │
│  │     Web UI          │     │         CLI                  │  │
│  │  (Qwik + Tailwind)  │     │   ml-engine <command>        │  │
│  │  Port 5173          │     │   30+ commands               │  │
│  └─────────┬───────────┘     └──────────┬───────────────────┘  │
│            │                            │                       │
│            └──────────┬─────────────────┘                       │
│                       ↓ HTTP                                    │
│  ┌──────────────────────────────────────────────────────────┐  │
│  │               REST API (Hono, Port 4321)                  │  │
│  │   65+ endpoints | Provider | Model | Config | Analytics  │  │
│  └──────────────────────────┬───────────────────────────────┘  │
│                             ↓                                   │
│  ┌──────────────────────────────────────────────────────────┐  │
│  │               @ml-engine/core                             │  │
│  │  ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐    │  │
│  │  │Provider  │ │  Model   │ │  Config  │ │Analytics │    │  │
│  │  │Registry  │ │ Catalog  │ │ Manager  │ │+ Budgets │    │  │
│  │  ├──────────┤ ├──────────┤ ├──────────┤ ├──────────┤    │  │
│  │  │ 10       │ │ Search   │ │ OpenCode │ │Usage     │    │  │
│  │  │providers │ │ Compare  │ │ Cursor   │ │Cost proj │    │  │
│  │  │Scanning  │ │ Aliases  │ │ Continue │ │Alerts    │    │  │
│  │  └──────────┘ └──────────┘ └──────────┘ └──────────┘    │  │
│  │  ┌──────────┐ ┌──────────┐ ┌──────────────────────┐     │  │
│  │  │Experiment│ │Orchestra-│ │ Utils                │     │  │
│  │  │ Tracker  │ │  tion    │ │ Env import | Export  │     │  │
│  │  └──────────┘ └──────────┘ └──────────────────────┘     │  │
│  └──────────────────────────┬───────────────────────────────┘  │
│                             ↓                                   │
│  ┌──────────────────────────────────────────────────────────┐  │
│  │               SQLite Store (bun:sqlite)                   │  │
│  │  ~/.local/share/ml-engine/engine.db                      │  │
│  │  Auto-migration | Tables: providers, models,             │  │
│  │  experiments, usage, budgets, settings                   │  │
│  └──────────────────────────────────────────────────────────┘  │
│                                                                 │
│  ┌──────────────────────────────────────────────────────────┐  │
│  │   @ai-lab/openai-compatible (forked AI SDK)              │  │
│  │   AI SDK v4 fork with minChunkSize SSE fix               │  │
│  │   Replaces legacy SSE proxy entirely                     │  │
│  └──────────────────────────────────────────────────────────┘  │
│                                                                 │
└─────────────────────────────────────────────────────────────────┘

---

Quick Start

# 1. Prerequisites
curl -fsSL https://bun.sh/install | bash
git clone https://github.com/ainjiner/ai-lab.git
cd ai-lab

# 2. Install
bun install

# 3. Build core engine
bun run build:core

# 4. Add your first provider
bun run cli provider add openai production YOUR_OPENAI_API_KEY

# 5. Scan available models
bun run cli provider scan openai-production

# 6. Sync to OpenCode (optional)
bun run cli config sync opencode

# 7. Start the API (port 4321)
bun run api

# 8. Start the Web UI (separate terminal, port 5173)
bun run dev:web

Open http://localhost:5173 to see the dashboard.

---

CLI Reference

ml-engine <command> [args]

Provider Commands

Command	Description
provider list	List all supported providers
provider instances	List configured instances
provider add <p> <name> <key>	Add a provider instance
provider remove <id>	Remove an instance
provider enable <id>	Enable an instance
provider disable <id>	Disable an instance
provider test <id>	Test API connection
provider scan <id>	Discover available models

Model Commands

Command	Description
model list [--provider <id>]	List all models
model search <query>	Search models by name
model info <id>	Get model details
model compare <id1> <id2>	Side-by-side comparison
model alias add <a> <mid> <pid>	Create model alias
model alias list	List all aliases
model alias remove <a>	Remove alias
model recommend [opts]	Get task-based recommendations

Config Commands

Command	Description
config list	List config targets
config sync <target>	Sync to target (opencode/cursor/continue/aider)
config sync --all	Sync all enabled targets
config preview <target>	Preview generated config
config enable <target>	Enable target
config disable <target>	Disable target

Settings Commands

Command	Description
settings show	Show current settings
settings set <key> <value>	Update a setting

Environment & Export

Command	Description
env detect	Detect provider env vars
env import	Import from PROVIDER_*_API_KEY
export	Show export summary
export all	Full JSON export to stdout
export save <path>	Save export to file

---

REST API

The API runs on http://localhost:4321/api.

Providers

Method	Endpoint	Description
GET	/providers	List all providers
GET	/providers/instances	List configured instances
POST	/providers/instances	Add instance
GET	/providers/instances/:id	Get instance
PATCH	/providers/instances/:id	Update instance
DELETE	/providers/instances/:id	Delete instance
POST	/providers/instances/:id/test	Test connection
POST	/providers/instances/:id/scan	Scan models

Models

Method	Endpoint	Description
GET	/models	List models (?provider=&search=)
GET	/models/compare?ids=a,b	Compare two models
GET	/models/aliases	List aliases
GET	/models/:id	Get model details
POST	/models/recommend	Get recommendations
POST	/models/aliases	Create alias
DELETE	/models/aliases/:alias	Delete alias

Config

Method	Endpoint	Description
GET	/config	Targets + settings
POST	/config/targets/:id/sync	Sync to target
POST	/config/targets/:id/toggle	Enable/disable target
POST	/config/sync-all	Sync all enabled
GET	/config/preview/:id	Preview config
GET	/config/settings	Get settings
POST	/config/settings	Update settings

Experiments

Method	Endpoint	Description
GET	/experiments	List experiments (?status=)
POST	/experiments	Create experiment
GET	/experiments/:id	Get experiment
POST	/experiments/:id/result	Save result
GET	/experiments/compare?ids=a,b	Compare experiments
DELETE	/experiments/:id	Delete experiment

Analytics

Method	Endpoint	Description
GET	/analytics/summary	Usage summary
GET	/analytics/breakdown	By provider/model
GET	/analytics/projection?days=30	Cost projection
GET	/analytics/export?format=json	Export usage data
GET	/budgets	List budgets
POST	/budgets	Create budget
GET	/budgets/check	Check alerts
DELETE	/budgets/:id	Delete budget

Orchestration

Method	Endpoint	Description
GET	/orchestration	Installed tools + OMO config
GET	/orchestration/agents	List agents
POST	/orchestration/agents	Save agent
GET	/orchestration/skills	List skills
GET	/orchestration/skills/:id	Get skill content

System

Method	Endpoint	Description
GET	/health	Health check
GET	/export	Export all config
POST	/import	Import config
POST	/import-env	Import from env vars
GET	/detect-env	Detect environment variables

---

Supported Providers

Provider	Type	Auth	Features
Baseten	Inference	API Key	tools, reasoning, JSON mode
OpenRouter	Gateway	API Key	vision, tools, reasoning
Together	Inference	API Key	vision, tools, prompt caching
Fireworks	Inference	API Key	vision, tools, prompt caching
Groq	Inference	API Key	vision, tools
Anthropic	Inference	API Key	vision, tools, reasoning, caching
OpenAI	Inference	API Key	vision, tools, reasoning, caching
Google	Inference	API Key	vision, tools
DeepSeek	Inference	API Key	tools, reasoning
Ollama	Local	None	vision

---

Programmatic Usage

import {
  providerRegistry,
  modelCatalog,
  configManager,
  experimentTracker,
  analyticsTracker,
  orchestrationManager,
  exportAll,
  importFromEnv,
} from "@ml-engine/core";

// Add a provider instance
const instance = providerRegistry.addInstance({
  providerId: "openai",
  name: "production",
  apiKey: "sk-...",
});

// Scan for models
const scan = await modelCatalog.scanModels(instance.id);

// Create an experiment
const exp = experimentTracker.create({
  name: "GPT-4o vs Claude 3.5 Sonnet",
  userPrompt: "Explain quantum entanglement simply",
  providerId: instance.id,
  modelId: "gpt-4o",
});

// Save a result
experimentTracker.saveResult(exp.id, {
  output: "Quantum entanglement is like...",
  tokensPrompt: 150,
  tokensCompletion: 75,
  latencyMs: 2400,
  costUsd: 0.0025,
});

// Sync config to OpenCode
configManager.syncToTarget("opencode");

// Export everything for backup or team sharing
const backup = exportAll();

// Auto-import from environment
const result = importFromEnv("PROVIDER_");

---

Storage

All data persists in SQLite at ~/.local/share/ml-engine/engine.db:

Table	Contents
providers	Provider registry metadata
instances	API keys, settings, scan results
models	Discovered models with pricing
aliases	Model alias mapping
experiments	Experiment configs and results
usage	Token usage records
budgets	Budget definitions and alerts
settings	Platform config preferences

---

Config Sync Targets

Target	Config Path	Auth Path
OpenCode	~/.config/opencode/opencode.json	~/.local/share/opencode/auth.json
Cursor	~/.cursor/config.json	—
Continue	~/.continue/config.json	—
Aider	~/.aider/config.json	—

---

Comparison with Other Tools

Feature	AI Lab	Helicone	Langfuse	LangSmith	W&B	Arize Phoenix
Self-Hosted	✅	✅	✅	❌	❌	✅
Open Source	✅ MIT	✅ Apache	✅ MIT	❌	❌	✅ MIT
Model Comparison	✅ Side-by-side	❌	❌	❌	❌	❌
Config Sync	✅ OpenCode-native	❌	❌	❌	❌	❌
Cost Projections	✅	✅	✅	✅	❌	❌
Budget Alerts	✅	❌	❌	❌	❌	❌
Experiment Tracking	✅	❌	✅	✅	✅	✅
Provider Registry	✅ 10 providers	❌	❌	❌	❌	❌
OMO Orchestration	✅	❌	❌	❌	❌	❌
Pricing	Free forever	$79+/mo	$59+/mo	$39/seat	$50+/mo	Free

---

Roadmap

Phase	Focus	Status
1. Foundation	Provider registry, model catalog, config sync, CLI, API, Web UI	✅ Complete
2. SSE Streaming	Forked AI SDK with minChunkSize fix	✅ Complete
3. RAG/CAG Support	Document chunking, retrieval eval, context analysis	🔜 Planned
4. Advanced Evaluation	LLM-as-judge, custom scorers, regression detection	🔜 Planned
5. Fine-tuning	Dataset curation from traces, training data export	🔜 Planned
6. Agent Development	Agent tracing, tool call visualization, multi-agent debug	🔜 Planned

---

Contributing

Contributions are welcome! See CONTRIBUTING.md for details.

Areas we'd love help with:

• 🔌 Additional provider integrations
• 🎨 Web UI improvements
• 📊 New analytics visualizations
• 📖 Documentation and tutorials
• 🧪 Test coverage

---

Part of the Ainjiner Ecosystem

AI Lab is built and maintained by Ainjiner — an open-source AI engineering organization from Indonesia.

---

License

MIT — Free forever, local-first, no restrictions.