🚀 Leaper Agent
The Agent Framework That Actually Remembers
Hermes-Compatible Agent Orchestration + DeepBrain 6-Layer Evolving Memory
🤔 The Problem
Every agent framework gives you tool-calling and multi-model support. None of them give you real memory.
- CrewAI? Flat key-value store, resets between runs.
- AutoGen? Conversation history only, no knowledge evolution.
- LangGraph? Checkpoints, not understanding.
- Hermes Agent? Skill files on disk — smart, but static.
Your agent forgets everything the moment the session ends.
💡 The Solution
Leaper Agent = Hermes Agent base + DeepBrain memory engine.
We took the proven Hermes Agent architecture — its agent loop, tool runtime, and multi-provider support — and integrated DeepBrain's 6-layer self-evolving memory engine. The result: an agent that gets smarter every time you talk to it.
Session 1: "I'm building a SaaS product with FastAPI and React."
Session 2: "What stack should I use for the admin panel?"
→ Agent recalls your tech choices, suggests FastAPI Admin + React
→ No manual context injection. It just knows.
🧠 Why Memory Matters
| What happens | Without memory | With DeepBrain |
|---|---|---|
| You mention your tech stack | Forgotten next session | Remembered forever, evolves over time |
| You correct the agent | Same mistake tomorrow | Learns the correction, never repeats |
| You have 50 conversations | Each starts from zero | Agent builds a knowledge graph of YOU |
| Knowledge conflicts | Silently contradicts itself | 4-Gate system detects & resolves conflicts |
6-Layer Memory Architecture
┌─────────────────────────────────────────┐
│ Layer 5: Meta-Knowledge │
│ "I know your tech preferences well, │
│ but I'm uncertain about your budget" │
├─────────────────────────────────────────┤
│ Layer 4: Archived — Historical ref │
├─────────────────────────────────────────┤
│ Layer 3: Consolidated — Cross-session │
├─────────────────────────────────────────┤
│ Layer 2: Long-Term — Validated facts │
├─────────────────────────────────────────┤
│ Layer 1: Short-Term — Recent context │
├─────────────────────────────────────────┤
│ Layer 0: Flash — Current session │
└─────────────────────────────────────────┘
↑ Auto-promotion via 4-Gate QC ↑
(Relevance · Novelty · Consistency · Utility)
This isn't RAG. This isn't vector search. This is knowledge that evolves — facts get validated, promoted, consolidated, and the agent develops meta-awareness of what it knows well and what it doesn't.
🚀 Quick Start
# Install
pip install leaper-agent
# Set your API key
export OPENAI_API_KEY=sk-xxx
# or: export ANTHROPIC_API_KEY=sk-ant-xxx
# or: export GEMINI_API_KEY=xxx
# Create an agent with the CEO Coach template
leaper create my-agent --template ceo-coach
# Start
leaper start my-agent
Deploy as Telegram Bot
leaper create my-bot --template ceo-coach --bot-token YOUR_TELEGRAM_TOKEN
leaper start my-bot
🏗️ Architecture
┌─────────────────────────────────────────┐
│ Leaper Agent Runtime │
│ (Hermes-compatible agent loop) │
├─────────┬──────────┬──────────┬────────┤
│ OpenAI │ Claude │ Gemini │ Ollama │
├─────────┴──────────┴──────────┴────────┤
│ Agent Orchestration │
│ Role templates · Tool calling │
├─────────────────────────────────────────┤
│ 🧠 DeepBrain Memory Engine │
│ 6-Layer · 4-Gate · Zero Dependencies │
│ SQLite-only · 100% Local · Evolving │
└─────────────────────────────────────────┘
Key difference from Hermes: Hermes stores skills as static Markdown files. Leaper Agent's DeepBrain automatically extracts, validates, and evolves knowledge from every conversation — no manual curation needed.
⚡ Hermes Compatibility
Leaper Agent is built on the Hermes Agent foundation. If you're familiar with Hermes, you'll feel at home:
- Same agent loop architecture
- Same multi-provider model support
- Same tool runtime
- Plus: 6-layer evolving memory that Hermes doesn't have
🔌 Supported Providers
| Provider | Models | Status |
|---|---|---|
| OpenAI | GPT-4o, GPT-4o-mini, o1/o3 | ✅ |
| Anthropic | Claude Sonnet, Opus, Haiku | ✅ |
| Gemini 2.0, 2.5 | ✅ | |
| Ollama | Any local model | ✅ |
📊 vs Other Frameworks
| Leaper Agent | Hermes | CrewAI | AutoGen | LangGraph | |
|---|---|---|---|---|---|
| Memory that evolves | ✅ 6-layer | ❌ Static files | ❌ Flat KV | ❌ Chat history | ❌ Checkpoints |
| Knowledge quality control | ✅ 4-Gate | ❌ | ❌ | ❌ | ❌ |
| Meta-knowledge | ✅ | ❌ | ❌ | ❌ | ❌ |
| Multi-provider | ✅ | ✅ | ✅ | ✅ | ✅ |
| Zero memory dependencies | ✅ SQLite only | N/A | ❌ Redis/Qdrant | ❌ | ❌ Vector DB |
| Hermes compatible | ✅ | ✅ | ❌ | ❌ | ❌ |
🇨🇳 China Version / 中国版
For Chinese LLM providers (DeepSeek, 通义千问, 智谱, Moonshot) + 150 role templates:
pip install leaper-agent-cn
🔗 Ecosystem
| Project | Description |
|---|---|
| OPC DeepBrain | Standalone memory engine (use in any framework) |
| OPC Agent | Local-first agent with Ollama |
| Leaper Agent CN | China-optimized with 150 templates |
📄 License
BSL-1.1 — Free for non-competitive use. Converts to Apache-2.0 after 4 years.
🤝 Contributing
We welcome contributions! See CONTRIBUTING.md.
📧 Contact: tech@deepleaper.com
Built with ❤️ by Deepleaper Technology / 跃盟科技
Other agents forget. Leaper Agent evolves.
Metadata
Release files for leaper-agent 2.0.3
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| leaper_agent-2.0.3.tar.gz | 5.9 MB | Without provenance | Details for leaper_agent-2.0.3.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| leaper_agent-2.0.3-py3-none-any.whl | Python 3 | none | any | Details for leaper_agent-2.0.3-py3-none-any.whl |
Total release size: 8.7 MB
Release files / leaper_agent-2.0.3.tar.gz
| Download URL | leaper_agent-2.0.3.tar.gz |
|---|---|
| Size | 5.9 MB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
916c31e0936e2f9e8b25cb9100ad9a23b2b36f09d5b2e1dcae02b7dcb778dc15
|
|
BLAKE2b-256 checksum How to use checksums |
ce4e5ed93a2ce10241f68ed47113308346e11ebe17b7f35f79b4cf169e48a2b7
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.2.0 CPython/3.12.10
|
Release files / leaper_agent-2.0.3-py3-none-any.whl
| Download URL | leaper_agent-2.0.3-py3-none-any.whl |
|---|---|
| Size | 2.8 MB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
56fcfaa730bb07fae962bf0c8fc5482265b0f265c7c39dbb2d3e2148c89e3284
|
|
BLAKE2b-256 checksum How to use checksums |
f745795e04327175215b4c056e1e52281c40a0d2de54d9b9e75c324435525669
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.2.0 CPython/3.12.10
|
Security for leaper-agent 2.0.3
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.