Skip to content

Commit 4ffd0e3

Browse files
author
Ismael Marchi
committed
feat: global infra upgrade + agent handshake layer (SKILL.md + llms.txt)
1 parent 6354399 commit 4ffd0e3

3 files changed

Lines changed: 191 additions & 10 deletions

File tree

‎README.md‎

Lines changed: 58 additions & 10 deletions
Original file line numberDiff line numberDiff line change
@@ -1,14 +1,19 @@
11
<div align="center">
22

3-
<h1>Synapse Layer — Long-Term Memory for AI Agents</h1>
3+
<h1>Synapse Layer</h1>
44

5-
<p><strong>Plug once. Remember forever. Zero-Knowledge. Zero-Amnesia. 🧠</strong></p>
5+
<p><strong>Persistent, Encrypted Memory Infrastructure for AI Agents</strong></p>
6+
7+
<p>Plug once. Remember forever. Zero-Knowledge. Zero-Amnesia. 🧠</p>
68

79
<br>
810

911
<a href="https://github.com/SynapseLayer/synapse-layer/actions"><img src="https://github.com/SynapseLayer/synapse-layer/actions/workflows/ci.yml/badge.svg" alt="CI"></a>
1012
<a href="https://smithery.ai/servers/synapselayer/synapse-protocol"><img src="https://smithery.ai/badge/synapselayer/synapse-protocol" alt="Synapse Layer on Smithery"></a>
1113
<a href="https://synapselayer.org/docs"><img src="https://img.shields.io/badge/Docs-Mintlify-0D9373" alt="Documentation"></a>
14+
<a href="https://pypi.org/project/synapse-layer/"><img src="https://img.shields.io/pypi/v/synapse-layer" alt="PyPI"></a>
15+
<a href="LICENSE"><img src="https://img.shields.io/badge/License-Apache_2.0-blue.svg" alt="License"></a>
16+
<a href="#-built-for-ai-agents"><img src="https://img.shields.io/badge/AI-Agent_Ready-blueviolet" alt="AI Agent Ready"></a>
1217

1318
<br><br>
1419

@@ -23,16 +28,16 @@
2328
```python
2429
from synapse_layer import SynapseMemory
2530

26-
memory = SynapseMemory(agent_id="my-agent")
31+
memory = SynapseMemory(agent_id="agent-1")
32+
33+
await memory.store(
34+
content="User prefers secure systems",
35+
confidence=0.95
36+
)
2737

28-
# 1. Agents save context automatically
29-
memory.save("User prefers absolute security and neural handover.")
38+
results = await memory.recall("user preferences")
3039

31-
# 2. Recall is deterministic and explainable
32-
info = memory.recall("user focus")
33-
# Result: "User prefers absolute security and neural handover."
34-
#
35-
# Every result includes a Trust Quotient (TQ) — a deterministic score based on:
40+
# Each result includes a Trust Quotient (TQ) — a deterministic score based on:
3641
# - Recency (how current the info is)
3742
# - Frequency (how often it was reinforced)
3843
# - Source Authority (how reliable the input source is)
@@ -56,6 +61,27 @@ Not as a feature — but as **infrastructure**.
5661

5762
---
5863

64+
## 🎯 Who Is This For
65+
66+
- **AI agent builders** — give your agents persistent context across sessions
67+
- **LLM infrastructure teams** — reduce token waste, increase reasoning consistency
68+
- **Production AI systems** — deterministic recall with zero hallucination risk
69+
- **Privacy-first organizations** — built-in LGPD/GDPR compliance with zero-knowledge architecture
70+
71+
---
72+
73+
## Before vs After
74+
75+
| | Without Memory | With Synapse Layer |
76+
|---|---|---|
77+
| **Session state** | Stateless — resets every turn | Persistent — survives across sessions |
78+
| **Token usage** | Reprocesses context every call | Up to 70% reduction via recall |
79+
| **Model switching** | Context lost between models | Signed handover (GPT-4 ↔ Claude) |
80+
| **Privacy** | Plaintext embeddings | AES-256-GCM + PII redaction + DP noise |
81+
| **Recall quality** | Non-deterministic | Deterministic, explainable, ranked by TQ |
82+
83+
---
84+
5985
## 🏢 Real-World Impact
6086

6187
### Used in production
@@ -217,6 +243,20 @@ Synapse Layer makes memory:
217243

218244
---
219245

246+
## 🤖 Built for AI Agents
247+
248+
This repository is optimized for autonomous agents, LLM orchestrators, and multi-agent systems.
249+
250+
| File | Purpose |
251+
|---|---|
252+
| [`SKILL.md`](SKILL.md) | Machine-readable agent interface specification |
253+
| [`llms.txt`](llms.txt) | AI crawler format for LLM discovery |
254+
| MCP native protocol | Direct integration via `forge.synapselayer.org/api/mcp` |
255+
256+
Synapse Layer is not just a library you call — it's **infrastructure your agents connect to**.
257+
258+
---
259+
220260
## 🎯 Roadmap & Community
221261

222262
| Version | Status | Highlights |
@@ -269,7 +309,15 @@ We are solving it — at the foundation level.
269309
---
270310

271311
<div align="center">
312+
313+
⭐ **[Star Synapse Layer](https://github.com/SynapseLayer/synapse-layer)** — Memory is not optional. Give your agents a past.
314+
315+
<br><br>
316+
272317
<strong>Giving Agents a Past. Giving Models a Soul. ⚗️</strong>
318+
273319
<br><br>
320+
274321
Built by <a href="https://synapselayer.org">Ismael Marchi</a> · <a href="https://x.com/synapselayer">@synapselayer</a>
322+
275323
</div>

‎SKILL.md‎

Lines changed: 74 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,74 @@
1+
# Synapse Layer Skill
2+
3+
Zero-Knowledge Memory Layer for AI Agents.
4+
5+
## Overview
6+
7+
Synapse Layer provides persistent, encrypted memory with deterministic recall and cross-session continuity.
8+
9+
- **MCP Endpoint:** `https://forge.synapselayer.org/api/mcp`
10+
- **SDK:** [github.com/SynapseLayer/synapse-layer](https://github.com/SynapseLayer/synapse-layer)
11+
- **PyPI:** [synapse-layer](https://pypi.org/project/synapse-layer/)
12+
13+
## Why Agents Should Use This
14+
15+
Agents are stateless by default. Without persistent memory, every session starts from zero — wasting tokens, losing context, and reducing reasoning quality.
16+
17+
Synapse Layer enables:
18+
- **Long-term memory** — context survives across sessions
19+
- **Trust-aware recall (TQ)** — deterministic, ranked results
20+
- **Cross-model continuity** — context transfers between GPT-4, Claude, and others
21+
22+
## Capabilities
23+
24+
| Tool | Function |
25+
|---|---|
26+
| `process_text` | Autonomous detection of decisions, milestones, and alerts |
27+
| `save_to_synapse` | Structured memory persistence with full security pipeline |
28+
| `backfill_embeddings` | Async vector embedding generation |
29+
| `health_check` | System health and capability report |
30+
31+
## Usage
32+
33+
```python
34+
from synapse_layer import SynapseMemory
35+
36+
memory = SynapseMemory(agent_id="agent-1")
37+
38+
memory.save("User prefers minimal interfaces")
39+
memory.recall("preferences")
40+
```
41+
42+
## Agent Rules
43+
44+
- Always pass `agent_id` to maintain identity across sessions
45+
- Prefer high TQ results when multiple memories match
46+
- Use `recall` before generating responses to leverage existing context
47+
- Let the security pipeline handle PII — do not pre-filter
48+
49+
## Security
50+
51+
- AES-256-GCM encryption at rest
52+
- 15+ PII redaction patterns (emails, phones, API keys, etc.)
53+
- Differential privacy on embeddings
54+
- Zero-knowledge architecture — no plaintext leaves the agent
55+
56+
## Integration
57+
58+
This skill connects to a remote MCP server. No local persistence required.
59+
60+
```json
61+
{
62+
"mcpServers": {
63+
"synapse-layer": {
64+
"url": "https://forge.synapselayer.org/api/mcp"
65+
}
66+
}
67+
}
68+
```
69+
70+
## Links
71+
72+
- [Documentation](https://synapselayer.org/docs)
73+
- [Forge Dashboard](https://synapselayer.org/forge)
74+
- [GitHub](https://github.com/SynapseLayer/synapse-layer)

‎llms.txt‎

Lines changed: 59 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,59 @@
1+
# Synapse Layer
2+
3+
> Persistent, encrypted memory infrastructure for AI agents.
4+
5+
## Overview
6+
7+
Synapse Layer is the missing memory primitive for AI systems. It provides persistent, encrypted, cross-model memory with deterministic recall — designed as infrastructure, not a feature.
8+
9+
## Core Concepts
10+
11+
- **agent_id**: Required identifier to maintain agent identity across sessions
12+
- **Memory persistence**: Cross-session, cross-model, encrypted at rest
13+
- **Trust Quotient (TQ)**: Deterministic reliability score for each recalled memory, based on recency, frequency, and source authority
14+
- **Neural Handover**: Signed context transfer between different AI models
15+
16+
## API
17+
18+
- **MCP endpoint**: `https://forge.synapselayer.org/api/mcp`
19+
- **Python SDK**: `pip install synapse-layer`
20+
- **Smithery**: `npx @smithery/cli install @synapselayer/synapse-protocol`
21+
22+
## Tools
23+
24+
| Tool | Description |
25+
|---|---|
26+
| process_text | Scans text for decisions, milestones, alerts — saves autonomously |
27+
| save_to_synapse | Direct structured memory persistence with full security pipeline |
28+
| backfill_embeddings | Async vector embedding generation for stored memories |
29+
| health_check | System health, version, capability report |
30+
31+
## Usage Rules
32+
33+
- Always recall before generating responses to leverage existing context
34+
- Prefer high TQ results when multiple memories are available
35+
- Use memory for personalization, continuity, and reasoning consistency
36+
- Let the security pipeline handle PII redaction automatically
37+
38+
## Security
39+
40+
- AES-256-GCM encryption
41+
- 15+ PII/secret redaction patterns
42+
- Differential privacy on embeddings
43+
- Zero-knowledge architecture
44+
45+
## Value
46+
47+
- Reduces token usage by up to 70%
48+
- Enables cross-session and cross-model continuity
49+
- Improves reasoning consistency with deterministic recall
50+
- Built-in LGPD/GDPR compliance
51+
52+
## Links
53+
54+
- Website: https://synapselayer.org
55+
- Documentation: https://synapselayer.org/docs
56+
- Forge Dashboard: https://synapselayer.org/forge
57+
- GitHub: https://github.com/SynapseLayer/synapse-layer
58+
- PyPI: https://pypi.org/project/synapse-layer/
59+
- Smithery: https://smithery.ai/servers/synapselayer/synapse-protocol

0 commit comments

Comments
 (0)