Open source · MIT licensed

The AI agent runtime that learns your workflow

Persistent 3-layer memory, 100+ model support, SkillBank auto-learning, and a full CLI — open source and production-ready today.

No spam. Unsubscribe any time.

✓ You're on the list — we'll be in touch!
100+
Models supported
3-layer
Persistent memory
330+
Tests passing
MIT
Open source
What it does

Everything you need to run agents in production

Not a toy. Not a wrapper. A runtime built for real, long-running workloads.

🧠
3-layer persistent memory
Master context (permanent facts), distilled long-term memory, and short-term episodic checkpoints — all in SQLite. Every session starts informed.
🔀
True multi-model routing
Claude, GPT-4o, DeepSeek, Gemini, Llama, Ollama — route by model, cost, or confidence score. Anthropic, OpenAI, and Google direct APIs supported alongside OpenRouter.
📚
SkillBank auto-learning
Successful task patterns are extracted, deduplicated, and injected into future sessions. Agents compound improvement across every run without any manual curation.
🪝
Hooks & MCP management
Pre/post-tool hooks for secrets scanning, type-checking, and audit logging. One-command MCP server setup with curated preset bundles: orager mcp add --preset recommended.
⚡
Profiles & auto-summarization
Switch agent personas with a single flag — --profile code-review, deploy, bug-fix. Sessions auto-compress at 70% token pressure so runs never die from context overflow.
🔬
OMLS training pipeline
Prompt tournament system with LLM-as-judge scoring, vision benchmarking, and burn-in validation. Fine-tune a LoRA adapter on your own trajectories and auto-route to it.
Quick start

Up and running in 60 seconds

# Install globally npm install -g orager # One-shot run — exits when done orager run "Summarise the last 10 git commits and open a PR" # Interactive chat — persistent memory, resumable sessions orager chat --model deepseek/deepseek-chat --profile code-review # Add a curated MCP server bundle in one command orager mcp add --preset recommended # Seed SkillBank from the community toolkit orager seed-toolkit --skills-only # Or use the library directly import { runAgentLoop } from "orager"; await runAgentLoop({ prompt: "Write and run tests for the auth module", model: "anthropic/claude-sonnet-4-5", apiKey: process.env.OPENROUTER_API_KEY, profile: "code-review", maxTurns: 30, onEmit: (e) => console.log(e), });
See it in action

Watch orager work

A real agent run — memory, tools, SkillBank, and all.

▶
3-minute walkthrough
Demo coming soon
Plans

Four ways to run OragerAI

The core runtime is MIT open source and always will be. Cloud services are coming soon.

Open Source
Free forever
Full runtime, every feature, self-hosted. MIT license — use it in production, embed it in your product, fork it.
  • 3-layer persistent memory
  • 100+ model support
  • SkillBank + hooks
  • Full CLI + library API
Available now
Orager Cloud
~$20 / mo
Managed agents with hosted memory and zero infrastructure. Connect your API key, point at your repo, run.
  • Everything in Open Source
  • Hosted SQLite memory
  • Web dashboard
  • Session history & replay
Coming soon
SkillBank Pro
~$30 / mo
Shared skill libraries across your whole team. Nightly OMLS training runs while you sleep. Wake up to a smarter agent.
  • Everything in Cloud
  • Team skill sharing
  • Nightly OMLS training
  • Prompt tournament dashboard
Coming soon
Enterprise
Custom
Self-hosted deployment, SSO, audit logs, SLA, private skill marketplace, and a dedicated support channel.
  • Everything in Pro
  • Private deployment
  • SSO + audit logs
  • Dedicated support
Contact us

Be first when cloud launches

One email when we're ready. No noise in between.

Unsubscribe any time.

✓ You're on the list — we'll be in touch!