Skip to content
View Grivn's full-sized avatar
👽
working from the Mars
👽
working from the Mars

Highlights

  • Pro

Organizations

@mnemon-dev

Block or report Grivn

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
Grivn/README.md

Hi, I'm Grivn 👋

Engineer by day, onchain by conviction.

I build distributed systems and tools for AI agents, with a background in BFT consensus and an interest in DeFi.

Currently building mnemon: persistent memory for AI agents, with graph-based recall and the host LLM as supervisor. I also build dsh-mnemon, composable memory for DeepSeek Harness.

Projects · Research · Background · Contributions · X / @grivn_eth


🔨 What I'm building

Project Description Stats
mnemon
Homepage
Persistent memory for AI agents in a single Go binary, with graph-based recall and host-LLM supervision. stars forks
dsh-mnemon Composable three-tier memory for DeepSeek Harness, with pluggable sources and strategies. stars forks
phalanx Byzantine fault-tolerant mempool with fair transaction ordering. stars forks
normalizejson Go package for normalizing JSON keys and values with templates. stars forks

📄 Research

Mnemon: Raw Records, Fast Judgments, Slow Thoughts

Paper · PDF · Code, prompts & run records

A memory agent that preserves raw, dated conversations and separates fast evidence judgments (System 1, Jev) from search planning and reasoning (System 2, LLM). It builds a compact evidence view when a question arrives, with a background index for broader conversation recall.

  • With GPT-4.1-mini: 91.7% on LoCoMo and 83.8% on LongMemEval-S, using under 4k context tokens per question.
  • With a reasoning model: 92.2% on LoCoMo and 94.4% on LongMemEval-S.
  • Scaling: on BEAM, per-question cost grows by just 1.11× as history grows from 100K to 10M tokens.

🧠 Background

  • 🎓 ZJU & WHU
  • 💼 Engineer @ ByteDance — AI agents · MCP · microservices · observability · AIOps · root cause analysis
  • ⛓️ Previously: consensus protocol R&D @ Hyperchain
  • 📄 Research: long-term memory for AI agents · fair ordering in BFT consensus protocols
  • ⚡ Tech: Go · Python · MCP

🤝 Contributions

Open-source projects I've contributed to
Project Description Stats
dsh-web-ui Plugin and skin collection for DeepSeek Harness Web UI — task board, Git graph, remote mobile UI, usage stats, and more stars forks
dsh-usage-stats Token usage heatmaps, per-model breakdowns, and DeepSeek account balance for DSH Web UI stars forks
CLIProxyAPI Wrap Gemini CLI, Claude Code, Codex, Qwen as OpenAI/Claude-compatible API stars forks
mcp-go Go implementation of the Model Context Protocol (MCP) stars forks
DIA Open-source financial data platform for digital assets stars forks
cometbft Byzantine fault-tolerant state machine replication engine (fork of Tendermint Core) stars forks
bamboo Source code for the ICDCS 2022 paper "Dissecting the Performance of Chained-BFT" stars forks
blockchain_conference_paper Curated list of blockchain academic papers by conference and year stars forks

🌐 Onchain

DeFi analyst | $HYPE · $PENDLE | Believer in the free movement of value

Pinned Loading

  1. mnemon-dev/mnemon mnemon-dev/mnemon Public

    LLM-supervised persistent memory for AI agents — graph-based recall, cross-session knowledge, single binary. Works with DeepSeek Harness, Claude Code, OpenClaw, and any agent runtime.

    Go 609 74

  2. omdsh-dev/dsh-mnemon omdsh-dev/dsh-mnemon Public

    Composable, view-based memory for DeepSeek Harness. Pluggable sources and strategies, with three-tier memory out of the box.

    TypeScript 446 37

  3. mnemon-memory-agent mnemon-memory-agent Public

    Long-term memory for AI agents on Jev. Keep raw records, judge them with a fast System 1 model and answer from under 4k tokens of context.

    TypeScript 4

  4. mark3labs/mcp-go mark3labs/mcp-go Public

    A Go implementation of the Model Context Protocol (MCP), enabling seamless integration between LLM applications and external data sources and tools.

    Go 9.2k 894

  5. router-for-me/CLIProxyAPI router-for-me/CLIProxyAPI Public

    Wrap Antigravity, ChatGPT Codex, Claude Code, Grok Build, Muse Code, Devin as an OpenAI/Gemini/Claude/Codex compatible API service, allowing you to enjoy the free Gemini Series, GPT Series, Grok Se…

    Go 54k 8.2k

  6. zhu1090093659/dsh-web zhu1090093659/dsh-web Public

    DeepSeek Harness (DSH) Web 插件聚合生态 · 万物皆插件,通过创意工坊分发||DeepSeek Harness (DSH) Web Plugin Aggregation Ecosystem · Everything is a plugin, distributed via the Creative Workshop

    TypeScript 8.3k 555