惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

月光博客
月光博客
小众软件
小众软件
爱范儿
爱范儿
Y
Y Combinator Blog
博客园 - Franky
美团技术团队
博客园 - 【当耐特】
The Cloudflare Blog
罗磊的独立博客
Hugging Face - Blog
Hugging Face - Blog
Jina AI
Jina AI
IT之家
IT之家
人人都是产品经理
人人都是产品经理
奇客Solidot–传递最新科技情报
奇客Solidot–传递最新科技情报
让小产品的独立变现更简单 - ezindie.com
让小产品的独立变现更简单 - ezindie.com
大猫的无限游戏
大猫的无限游戏
Apple Machine Learning Research
Apple Machine Learning Research
博客园 - 聂微东
WordPress大学
WordPress大学
V
Visual Studio Blog
博客园_首页
阮一峰的网络日志
阮一峰的网络日志
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻
有赞技术团队
有赞技术团队

Hacker News - Newest: "AI"

AI can't read an investor deck AI as an attorney? Student uses ChatGPT, Gemini to sue UW over alleged racial discrimination Hacking MCP Servers in AI Systems – The Rug Pull: Tool Changes After Approval GitHub - MeepCastana/KubeezCut: Free Web based video editor Can AI judge journalism? A Thiel-backed startup says yes, even if it risks chilling whistleblowers Coming soon: 10 Things That Matter in AI Right Now DARPA built an AI to fact-check enemy weapons claims What explains heterogeneity in AI adoption? When AI Meets Muscle: Context-Aware Electrical Stimulation Promises a New Way to Guide Human Movements - Department of Computer Science AI Changed How We Build. It Did Not Change What Matters. Linux rules on using AI-generated code - Copilot is OK, but humans must take 'full responsibility for the… Meta spins up AI version of Mark Zuckerberg to engage with employees Code Mode: Let Your AI Write Programs, Not Just Call Tools | TanStack Blog GitHub - Delavalom/graft: Go framework for building AI agents. Type-safe tools, multi-provider (OpenAI, Anthropic, Gemini, Bedrock), zero vendor SDKs. India's TCS tops estimates, says new AI models did not dent services demand Gen Z's fading AI hype Strong feeling: we are in a folded AI reality GitHub - machinarii/total-recall-catalog: A reference catalog of latest knowledge retrieval, memory & RAG systems GitHub - mensfeld/code-on-incus: Give each AI agent its own isolated machine with root, Docker, and systemd. Active defense detects and stops threats automatically.. Quantization, LoRA, and the 8% Problem: Benchmarking Local LLMs for Production AI Iran war: We spoke to the man making Lego-style AI videos that experts say are powerful propaganda Powell, Bessent discussed Anthropic's Mythos AI cyber threat with major U.S. banks GitHub - immartian/bellamem: Persistent belief-graph memory for AI agents. Retrieves decisive context by importance — not recency, not RAG, not /compact. recursive-mode: The Repo-Native Operating System for AI Engineering After the attack on Sam Altman's home, will AI CEO's go on the offensive? The biggest advance in AI since the LLM Opus 4.6 vs GPT 5.4 One Prompt Unity World Generation Test “AI polls” are fake polls Client Challenge Can AI be a 'child of God'? Inside Anthropic's meeting with Christian leaders
GitHub - sovseal/core: Private Memory Layer for your AI A...
radebe49 · 2026-05-25 · via Hacker News - Newest: "AI"

sovseal — private memory for AI agents

Private memory layer for AI agents.
Stop losing context across sessions. Portable, local-first, and zero data leaks.

@sovseal/mcp-server · @sovseal on npm · Quickstart · Changelog

Npm package License: Apache 2.0 Node Supabase

Measured Performance & Benchmarks (May 2026)

Cloud-hosted memory layers force a tradeoff between latency, privacy, and cost. Every recall is a 200–800 ms round-trip to someone else's database. sovseal collapses the tradeoff. LanceDB and Transformers.js run inside your agent's process.

Workload Operation p50 p95 p99 Network
10K records · 1K queries recall_memory (warm) 6.1 ms 10.4 ms 21.8 ms 0 RTT
Cold start recall_memory (first call) ~1.2 s 0 RTT
Single write store_memory 3.8 ms 7.2 ms 12.5 ms 0 RTT (write-behind)

All benchmarks reproduce with: pnpm --filter @sovseal/mcp-server test bench-v2 (10K pre-seeded memories, 1K sequential queries, CPU-bound ONNX embeddings on commodity hardware).

What makes it fast and secure:

  • Sub-25 ms p99 recall — semantic search is a local vector query, not an HTTP call.
  • Zero-knowledge by construction — AES-256-GCM encryption before leaving the device.
  • Verified Semantic Recall (VSR) — every load re-derives sha256(canonicalize(payload)) and fails closed on mismatch.
  • Deterministic lineage — snapshot graph enables byte-equal state restoration.

Introduction

sovseal gives autonomous agents and AI assistants persistent, portable context that you actually own. It drops into any MCP-compatible client and stops your agents from forgetting workflows across sessions. Built for true portability and privacy, it guarantees zero data leaks by running entirely local-first with zero-knowledge cloud sync.

Key Features & Use Cases

Core Capabilities:

  • Local-First Semantic Memory: On-device LanceDB + 384-dim Transformers.js embeddings for 0-RTT recall.
  • Zero-Knowledge Architecture: The server only sees ciphertext. End-to-end AES-256-GCM.
  • Write-Behind Replication: Tool calls return on local commit; ciphertext sync happens asynchronously.
  • Developer-Friendly: Drop-in MCP server, Node SDK, and self-hosted edge endpoints.

Applications:

  • AI Coding Assistants: Claude Desktop, Cursor, and Windsurf need persistent, private memory across long sessions.
  • Agent Frameworks: ElizaOS, Hermes, CrewAI, and LangGraph natively consume MCP.
  • Privacy-Sensitive Teams: Healthcare, legal, and defense teams that require plaintext to stay on-device.

🚀 Quickstart Guide

Choose your path

sovseal exposes one protocol with three delivery shapes. Pick by where your code runs:

MCP Server Node SDK Self-Hosted Edge
Best for Any MCP-compatible client or agent framework In-process import inside a Node/TS service Full data residency on infra you control
Install npx -y @sovseal/mcp-server npm install @sovseal/sdk supabase functions deploy v2-agent-state
Transport stdio · HTTP · SSE Direct function calls HTTPS to your endpoint
Auth Self-asserting (sov_proj_<uuid>) API key or self-asserting Bring your own
Recall latency 0 RTT (local LanceDB) 0 RTT (local LanceDB) 0 RTT (local LanceDB)
Cost Free Free Your Supabase bill

MCP Server

The default for AI coding assistants and any agent framework that speaks MCP. One server binary, one config snippet, every client.

Claude Desktop, Cursor, Windsurf, Zed — add to mcp.json or claude_desktop_config.json:

{
  "mcpServers": {
    "sovseal-memory": {
      "command": "npx",
      "args": ["-y", "@sovseal/mcp-server"]
    }
  }
}

One-line install for Claude Code:

claude mcp add sovseal -- npx -y @sovseal/mcp-server

System Prompt Snippet (Claude Desktop / Cursor Project Instructions): To ensure the AI natively uses the memory server, paste this into your custom instructions:

This session has persistent memory via sovseal. At conversation start, check the sovseal://context/recent resource. Whenever the user shares preferences, plans, or personal context, store it via store_memory.

For always-on autonomous agents, switch to HTTP/SSE transport so the sovseal process outlives a single tool invocation:

SOVSEAL_TRANSPORT=sse SOVSEAL_PORT=4040 npx -y @sovseal/mcp-server

Node SDK (Library)

When you're building a backend service and want to manage agent state snapshot persistence programmatically, import the Node SDK client. It manages end-to-end AES-256-GCM encryption client-side and replicates checkpoints to your persistence cloud tier.

import { AgentStateClient, CryptoService } from "@sovseal/sdk";

const client = new AgentStateClient({
  endpoint: "https://your-project.supabase.co/functions/v1/v2-agent-state",
  apiKey: "sov_proj_your_project_uuid",
});

const key = await CryptoService.generateAESKey();

// Client-side AES-GCM encrypted snapshot upload
const receipt = await client.snapshot({
  key,
  payload: {
    agent_id: "agent_a1b2c3",
    sequence_number: 0,
    parent_snapshot: null,
    policy_hash: "0000000000000000000000000000000000000000000000000000000000000000",
    timestamp: new Date().toISOString(),
    wallet_balances: { USDC: { "8453": "50000000" } }, // $50 in 6 decimals
    active_context: {
      preference: "Customer prefers wire transfers over ACH for >$50k settlements."
    }
  }
});

// Restore the latest state snapshot
const { receipt: latestReceipt, ciphertextUrl } = await client.restore({
  agentId: "agent_a1b2c3",
});

Self-Hosted Edge

The replication endpoint is a Deno edge function — open source, deployable to any Supabase project for full data-residency control.

# From the repo root
supabase functions deploy v2-agent-state
supabase db push    # applies the agent_state_snapshots schema

# Point any client at it:
export SOVSEAL_ENDPOINT="https://<your-project>.supabase.co/functions/v1/v2-agent-state"

Source: supabase/functions/v2-agent-state/.

Basic Usage

Drop-in patterns for an agent loop using the @sovseal/sdk:

import OpenAI from "openai";
import { AgentStateClient, CryptoService, decryptJson } from "@sovseal/sdk";

const openai = new OpenAI();
const client = new AgentStateClient({
  endpoint: "https://your-project.supabase.co/functions/v1/v2-agent-state",
  apiKey: "sov_proj_your_project_uuid",
});
const key = await CryptoService.generateAESKey();

async function chatWithMemory(agentId: string, message: string) {
  let context = "";
  try {
    // Restore latest state snapshot
    const { receipt, ciphertextUrl } = await client.restore({ agentId });
    const res = await fetch(ciphertextUrl);
    const encryptedBytes = new Uint8Array(await res.arrayBuffer());
    
    // Decrypt the ciphertext client-side using the local AES key
    const payload = await decryptJson(encryptedBytes, key);
    context = JSON.stringify(payload.active_context);
  } catch (err) {
    console.log("No previous state snapshot found or failed to restore");
  }

  const reply = await openai.chat.completions.create({
    model: "gpt-5-mini",
    messages: [
      { role: "system", content: `You are a helpful assistant.\nContext:\n${context}` },
      { role: "user", content: message },
    ],
  });

  const assistant = reply.choices[0].message.content ?? "";

  // Save the updated state snapshot
  await client.snapshot({
    key,
    payload: {
      agent_id: agentId,
      sequence_number: 1, // Incremented in a production agent loop
      parent_snapshot: null,
      policy_hash: "0000000000000000000000000000000000000000000000000000000000000000",
      timestamp: new Date().toISOString(),
      wallet_balances: {},
      active_context: { lastMessage: message, reply: assistant }
    }
  });
  
  return assistant;
}

🔗 Integrations & Agent Frameworks

All of these consume the same MCP server through their first-class MCP support. No sovseal-specific adapter is required:

Framework How it consumes MCP
ElizaOS @fleek-platform/eliza-plugin-mcp (config-only, stdio or SSE)
Hermes Agent Native MCP, stdio + remote HTTP
CrewAI mcps=[...] field on the agent
LangGraph · LangChain langchain-mcp-adapters (npm + PyPI)
Microsoft Agent Framework Built-in MCP workbench
OpenAI Agents SDK Native MCPServerStdio / MCPServerSse
OpenClaw Native openclaw mcp consumer

🛠️ MCP Tools Exposed to the LLM

Tool Args Behavior
store_memory { content: string } Embed (384-dim, on-device) → write to local LanceDB → return. Ciphertext replication runs write-behind; nothing blocks.
recall_memory { query: string, topK?: number } Embed query (LRU-cached) → vector search local LanceDB → return top-K matches ranked by L2 distance. 0 RTT.

🔒 Threat Model — Read this before depending on it

  • Confidentiality. AES-256-GCM with a 96-bit random IV per snapshot. The server cannot read your context.
  • Integrity (VSR). Every recall re-derives sha256(canonicalize(payload)) and compares against the stored client_payload_hash. Corrupted storage fails closed.
  • Authentication. Bearer token = sov_proj_<uuid v4> in ~/.sovseal/config.json.
  • Storage. Ciphertext lands in a Supabase Storage bucket. Object paths are SHA-256-derived and unguessable without your project_id.
  • Loss. Lose ~/.sovseal/config.json → lose every snapshot ever made. There is no escrow, no recovery flow. Back it up.

🤝 Contributing

pnpm install
pnpm --filter @sovseal/mcp-server build
node packages/sovseal-mcp-server/dist/index.js     # stdio MCP server

Test (unit + integration + crypto round-trip):

pnpm --filter @sovseal/mcp-server test
pnpm --filter @sovseal/mcp-server test bench-v2

See CONTRIBUTING.md.

🗺️ Roadmap

The core open-source modules — the stdio/SSE MCP server, Node SDK, and self-hosted Supabase edge functions — are fully production-ready, free forever, and licensed under Apache 2.0.

Upcoming capabilities on our public roadmap:

  • Hosted sync tier — managed, high-availability replication for teams that don't want to run their own edge function.
  • Wallet-key mode — opt-in Arweave permanence via Base L2 and Irys.
  • Native framework adapters — specialized adapters for environments where MCP isn't a direct fit.

Track progress in CHANGELOG.md and the issue tracker.

📚 Documentation & Support

⚖️ License

Apache 2.0 — for the entire public surface (MCP server, SDK, core protocol, edge function, docs). Fork it, ship it, run it however you like.

Built by the sovseal team — local-first, zero-knowledge, free.