[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-84321":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":9,"totalLinesOfCode":9,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":9,"subscribersCount":16,"size":16,"stars1d":17,"stars7d":17,"stars30d":17,"stars90d":16,"forks30d":16,"starsTrendScore":18,"compositeScore":19,"rankGlobal":9,"rankLanguage":9,"license":9,"archived":20,"fork":20,"defaultBranch":21,"hasWiki":20,"hasPages":20,"topics":22,"createdAt":9,"pushedAt":9,"updatedAt":41,"readmeContent":42,"aiSummary":9,"trendingCount":16,"starSnapshotCount":16,"syncStatus":43,"lastSyncTime":44,"discoverSource":45},84321,"hivemind","activeloopai\u002Fhivemind","activeloopai","One brain for all your agents",null,"https:\u002F\u002Fgithub.com\u002Factiveloopai\u002Fhivemind","TypeScript",758,48,7,30,0,134,402,9.07,false,"main",[23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40],"ai","ai-agents","ai-memory","anthropic","artificial-intelligence","claude","claude-agent-sdk","claude-agents","claude-code-plugin","claude-skills","codex","embeddings","long-term-memory","memory-engine","openclaw","openclaw-skills","postgres","rag","2026-06-12 02:04:39","\u003Ch1 align=\"center\">\n  \u003Cbr>\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Factiveloopai\u002Fhivemind\">\n    \u003Cimg src=\"https:\u002F\u002Fraw.githubusercontent.com\u002Factiveloopai\u002Fhivemind\u002Fmain\u002Fdocs\u002Fpublic\u002Fhivemind-logo.svg\" alt=\"Hivemind\" width=\"120\">\n  \u003C\u002Fa>\n  \u003Cbr>\n  Hivemind\n  \u003Cbr>\n\u003C\u002Fh1>\n\n\u003Ch4 align=\"center\">One brain for all your agents\u003C\u002Fh4>\n\n\u003Cp align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Fwww.npmjs.com\u002Fpackage\u002F@deeplake\u002Fhivemind\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fnpm\u002Fv\u002F@deeplake\u002Fhivemind?color=blue&label=npm&style=for-the-badge\" alt=\"npm\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Factiveloopai\u002Fhivemind\u002Fstargazers\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Factiveloopai\u002Fhivemind?style=for-the-badge&logo=github\" alt=\"GitHub stars\">\u003C\u002Fa>\n  \u003Ca href=\"LICENSE\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLicense-Apache%202.0-blue.svg?style=for-the-badge\" alt=\"License\">\u003C\u002Fa>\n  \u003Ca href=\"package.json\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fnode-%3E%3D22.0.0-brightgreen.svg?style=for-the-badge\" alt=\"Node\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fdeeplake.ai\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPowered%20by-Deeplake-orange.svg?style=for-the-badge\" alt=\"Deeplake\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fwww.ycombinator.com\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FY%20Combinator-backed-ff6600.svg?style=for-the-badge\" alt=\"Y Combinator backed\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fdiscord.gg\u002FEeGjnyDBx\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FDiscord-chat-5865F2?logo=discord&logoColor=white&style=for-the-badge\" alt=\"Join us on Discord\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fjoin.slack.com\u002Ft\u002Fhubdb\u002Fshared_invite\u002Fzt-35zr0yil0-lnzJcQhACsBlB7~3lufrCg\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FSlack-chat-4A154B?logo=slack&logoColor=white&style=for-the-badge\" alt=\"Join us on Slack\">\u003C\u002Fa>\n\u003C\u002Fp>\n\n\u003Cp align=\"center\">\n  Auto-learning, cloud-backed shared brain for \u003Cb>Claude Code • OpenClaw • Codex • Cursor • Hermes • pi\u003C\u002Fb> agents.\u003Cbr>\n\u003C\u002Fp>\n\n> One engineer's agent figures out a tricky migration on Monday.\n>\n> Tuesday, every agent on the team can execute the pattern.\n\nOn [LoCoMo](https:\u002F\u002Farxiv.org\u002Fabs\u002F2402.17753), the public long-context memory benchmark, Hivemind is **25% cheaper, 1.7× fewer tokens, and 31% fewer turns** than running without shared memory. ([See the numbers below.](#benchmarks))\n\n**Beyond memory.** Hivemind doesn't just remember. It mines your team's traces for repeated patterns and codifies them into reusable skills that propagate back into every agent on the team. The agent your junior engineer used this morning is sharper because of what your senior engineer's agent figured out last week.\n\n- 📥 **Captures** every session's prompts, tool calls, and responses as structured traces in Deeplake\n- 🧠 **Codifies** patterns into reusable `SKILL.md` files, available to every agent on your team\n- 🔍 **Searches** traces and skills with hybrid lexical + semantic retrieval (BM25 fallback when embeddings off)\n- 🔗 **Propagates** capability across sessions, agents, teammates, and machines in real time\n- 📁 **Intercepts** file operations on `~\u002F.deeplake\u002Fmemory\u002F` through a virtual filesystem backed by SQL\n- 📝 **Summarizes** sessions into AI-generated wiki pages via a background worker at session end\n- ☁️ **BYOC**: keep data in your own GCS, Azure, S3, or on-prem bucket. [See Security & storage](#security--storage)\n\n## Benchmarks\n\nOn the [LoCoMo](https:\u002F\u002Farxiv.org\u002Fabs\u002F2402.17753) long-context memory benchmark (100 QA pairs, Claude Haiku via `claude -p`, hybrid lexical + semantic retrieval), Hivemind cuts cost, tokens, and turns versus a no-memory baseline:\n\n| Metric            | Baseline | Hivemind | Improvement      |\n|-------------------|----------|----------|------------------|\n| Cost \u002F 100 QA     | $8.94    | $6.65    | **25% cheaper**  |\n| Tokens \u002F question | 1,700    | 1,008    | **1.7× fewer**   |\n| Turns \u002F question  | 8.9      | 6.2      | **31% fewer**    |\n\nThe agent reaches the answer in fewer turns with less context, because the prior work is already in scope at recall time, not re-derived per session.\n\n## Quick start\n\nOne command, all your agents:\n\n```bash\nnpm install -g @deeplake\u002Fhivemind && hivemind install\n```\n\nThe installer detects every supported assistant on your machine (table below), wires up the hooks, and shows a one-line consent prompt before opening a browser for sign-in. Restart your assistants after install.\n\n**Headless \u002F CI installs:** pass an API token instead of using the browser flow:\n\n```bash\nHIVEMIND_TOKEN=\u003Cyour-token> hivemind install\n# or\nhivemind install --token \u003Cyour-token>\n```\n\nGet a token from your account settings on https:\u002F\u002Fdeeplake.ai. With no token in a non-interactive shell, the install completes with hooks but skips sign-in; run `hivemind login` later to enable shared memory.\n\n**Install for a specific assistant only:**\n\n```bash\nhivemind install --only claude\nhivemind claude install    # equivalent\nhivemind codex install\nhivemind claw install\nhivemind cursor install\nhivemind hermes install\nhivemind pi install\n```\n\n**Check what's wired up:**\n\n```bash\nhivemind status\n```\n\n**Supported assistants:**\n\n| Platform         | Integration                                      | Auto-capture | Auto-recall |\n|------------------|--------------------------------------------------|--------------|-------------|\n| **Claude Code**  | Marketplace plugin                               | ✅           | ✅          |\n| **OpenClaw**     | Native extension                                 | ✅           | ✅          |\n| **Codex**        | Hooks (`hooks.json`)                             | ✅           | ✅          |\n| **Cursor**       | Hooks (`hooks.json` 1.7+)                        | ✅           | ✅          |\n| **Hermes Agent** | Shell hooks (`config.yaml`) + skill + MCP server | ✅           | ✅          |\n| **pi**           | Extension API (`pi.on(...)`) + skill + AGENTS.md | ✅           | ✅          |\n\n### Alternative install paths\n\n\u003Cdetails>\n  \u003Csummary>\u003Cb>Claude Code plugin marketplace\u003C\u002Fb>\u003C\u002Fsummary>\n\nIf you prefer Claude Code's native plugin marketplace:\n\n```\n\u002Fplugin marketplace add activeloopai\u002Fhivemind\n\u002Fplugin install hivemind\n\u002Freload-plugins\n\u002Fhivemind:login\n```\n\nAuto-updates on each session start. Manual update: `\u002Fhivemind:update`.\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n  \u003Csummary>\u003Cb>OpenClaw ClawHub\u003C\u002Fb>\u003C\u002Fsummary>\n\n```\nopenclaw plugins install clawhub:hivemind\n```\n\nThen type `\u002Fhivemind_login` in chat, click the auth link, and sign in.\n\n#### Commands\n\n| Command | Description |\n|---------|-------------|\n| `\u002Fhivemind_login` | Sign in via device flow |\n| `\u002Fhivemind_capture` | Toggle capture on\u002Foff |\n| `\u002Fhivemind_whoami` | Show current org and workspace |\n| `\u002Fhivemind_orgs` | List organizations |\n| `\u002Fhivemind_switch_org \u003Cname>` | Switch organization |\n| `\u002Fhivemind_workspaces` | List workspaces |\n| `\u002Fhivemind_switch_workspace \u003Cid>` | Switch workspace |\n| `\u002Fhivemind_update` | Check for plugin updates |\n\nAuto-recall and auto-capture are enabled by default. Data is stored in the same `sessions` table as Claude Code and Codex.\n\n#### Coexistence with `memory-core`\n\nHivemind runs **alongside** OpenClaw's built-in `memory-core` plugin. It does **not** claim the memory slot, so `memory-core`'s dreaming cron (`\"0 3 * * *\"`) and other memory-slot-dependent jobs keep working. Hivemind captures session activity and exposes its own commands; `memory-core` keeps owning recall\u002Fpromotion\u002Fdreaming.\n\n#### Troubleshooting\n\n- **Hivemind seems slow or unresponsive.** Check the agent model in `~\u002F.openclaw\u002Fopenclaw.json` under `agents.defaults.model`. Hivemind makes many small tool calls per turn; a large reasoning model like Opus will feel sluggish. Recommended default: `anthropic\u002Fclaude-haiku-4-5-20251001`.\n- **`openclaw model \u003Cid>` says \"plugins.allow excludes model\".** The `model` plugin CLI is disabled by default. Edit `~\u002F.openclaw\u002Fopenclaw.json` directly (key `agents.defaults.model`) and restart the gateway: `systemctl --user restart openclaw-gateway.service`.\n- **Model switch rejected as \"not allowed\".** Use the exact dated provider-prefixed ID (`anthropic\u002Fclaude-haiku-4-5-20251001`, `anthropic\u002Fclaude-sonnet-4-6`). Legacy IDs like `claude-3-5-haiku-latest` and unprefixed bare IDs are not on OpenClaw's allowlist.\n- **Self-update via Telegram fails with \"elevated is not available\".** `tools.elevated.allowFrom` must include `telegram` before elevated commands work from that channel. Safer alternative: run the upgrade in a local shell with `openclaw plugins update hivemind`.\n- **`npm error EACCES` during self-update.** OpenClaw was installed under a root-owned npm prefix (e.g. `\u002Fusr\u002Flib\u002Fnode_modules\u002Fopenclaw`). Reinstall under a user-writable prefix, or run the update with appropriate privileges locally, not via a channel.\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n  \u003Csummary>\u003Cb>Codex (manual)\u003C\u002Fb>\u003C\u002Fsummary>\n\nTell Codex to fetch and follow the install instructions:\n\n```\nFetch and follow instructions from https:\u002F\u002Fraw.githubusercontent.com\u002Factiveloopai\u002Fhivemind\u002Fmain\u002Fcodex\u002FINSTALL.md\n```\n\nOr run the installer script directly:\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002Factiveloopai\u002Fhivemind.git ~\u002F.codex\u002Fhivemind\n~\u002F.codex\u002Fhivemind\u002Fcodex\u002Finstall.sh\n```\n\nRestart Codex to activate.\n\n**First launch — trust the hooks.** Codex shows a **\"Hooks need review\"** prompt before it will run hivemind's hooks:\n\n```text\nHooks need review\n2 hooks are new or changed.\nHooks can run outside the sandbox after you trust them.\n\n   1. Review hooks\n › 2. Trust all and continue\n   3. Continue without trusting (hooks won't run)\n```\n\nChoose **`2. Trust all and continue`** — otherwise the hooks won't run and hivemind stays inactive.\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n  \u003Csummary>\u003Cb>Cursor (1.7+)\u003C\u002Fb>\u003C\u002Fsummary>\n\nThe unified installer wires six lifecycle events in `~\u002F.cursor\u002Fhooks.json`: sessionStart, beforeSubmitPrompt, postToolUse, afterAgentResponse, stop, sessionEnd. Hooks fork a Node bundle at `~\u002F.cursor\u002Fhivemind\u002Fbundle\u002F` per event. Restart Cursor after install to load.\n\n```bash\nhivemind cursor install\n```\n\nAuto-capture is enabled the same way as Claude Code \u002F Codex \u002F OpenClaw.\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n  \u003Csummary>\u003Cb>Hermes Agent\u003C\u002Fb>\u003C\u002Fsummary>\n\nWires shell hooks into `~\u002F.hermes\u002Fconfig.yaml` (`pre_llm_call` \u002F `post_tool_call` \u002F `post_llm_call` \u002F `on_session_end`) for auto-capture, drops the bundle at `~\u002F.hermes\u002Fhivemind\u002Fbundle\u002F`, registers the shared MCP server (`~\u002F.hivemind\u002Fmcp\u002Fserver.js`) under `mcp_servers.hivemind`, and installs an `agentskills.io`-compatible skill at `~\u002F.hermes\u002Fskills\u002Fhivemind-memory\u002F` for recall.\n\n```bash\nhivemind hermes install\n```\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n  \u003Csummary>\u003Cb>pi (badlogic\u002Fpi-mono coding-agent)\u003C\u002Fb>\u003C\u002Fsummary>\n\nUpserts an idempotent BEGIN\u002FEND marker block into `~\u002F.pi\u002Fagent\u002FAGENTS.md` (auto-loaded every turn) and drops a TypeScript extension at `~\u002F.pi\u002Fagent\u002Fextensions\u002Fhivemind.ts`. The extension subscribes to pi's lifecycle events (`session_start` \u002F `input` \u002F `tool_result` \u002F `message_end`) for auto-capture and registers `hivemind_search`, `hivemind_read`, `hivemind_index` as first-class pi tools.\n\n```bash\nhivemind pi install\n```\n\nNote: no per-agent SKILL.md is dropped under `~\u002F.pi\u002Fagent\u002Fskills\u002F`; pi reads skills from both that directory AND the shared `~\u002F.agents\u002Fskills\u002F` location. If the codex installer has run on the same machine, pi picks up the hivemind skill from the shared `~\u002F.agents\u002Fskills\u002Fhivemind-memory` symlink automatically. The AGENTS.md block plus the registered tools cover the action surface in either case.\n\u003C\u002Fdetails>\n\n\n### Uninstall\n\n```bash\nhivemind uninstall              # remove from every detected assistant\nhivemind codex uninstall        # remove from one\n```\n\n## How it works\n\n**Capture → Codify → Propagate → Compound.** Every coding-agent interaction (prompt, tool call, response) is captured as a structured trace in Deeplake. A background worker mines traces for repeated patterns and codifies them into `SKILL.md` files, scoped to your workspace. Codified skills propagate into every Hivemind-connected agent's context at inference time. The agent your junior engineer used this morning is sharper because of what your senior engineer's agent figured out last week.\n\n## Features\n\n### 🔍 Natural search\n\nJust ask your agent naturally:\n\n```\n\"What was Emanuele working on?\"\n\"Search traces for authentication bugs we've solved\"\n\"What did we decide about the API design?\"\n\"Show me skills my team has codified for handling migrations\"\n```\n\n### 🔒 Privacy controls\n\nDisable capture entirely:\n\n```bash\nHIVEMIND_CAPTURE=false claude\n```\n\nEnable debug logging:\n\n```bash\nHIVEMIND_DEBUG=1 claude\n```\n\n## ⚠️ Data collection notice\n\nThis plugin captures session activity and stores it in your Deeplake workspace:\n\n| Data                  | What's captured                    |\n|-----------------------|------------------------------------|\n| User prompts          | Every message you send             |\n| Tool calls            | Tool name + full input             |\n| Tool responses        | Full tool output                   |\n| Assistant responses   | The agent's final response         |\n| Subagent activity     | Subagent tool calls and responses  |\n| Codified skills       | Patterns extracted from traces     |\n\n**All users in your Deeplake workspace can read this data.** That's the design. Shared capability requires shared substrate. A DATA NOTICE is displayed at the start of every session. Workspace-level isolation prevents data leakage between orgs.\n\n## Configuration\n\n| Variable                  | Default                   | Description                                |\n|---------------------------|---------------------------|--------------------------------------------|\n| `HIVEMIND_TOKEN`          | _(none)_                  | API token (auto-set by login)              |\n| `HIVEMIND_ORG_ID`         | _(none)_                  | Organization ID (auto-set by login)        |\n| `HIVEMIND_WORKSPACE_ID`   | `default`                 | Workspace name                             |\n| `HIVEMIND_API_URL`        | `https:\u002F\u002Fapi.deeplake.ai` | API endpoint                               |\n| `HIVEMIND_TABLE`          | `memory`                  | SQL table for summaries and virtual FS     |\n| `HIVEMIND_SESSIONS_TABLE` | `sessions`                | SQL table for per-event session capture    |\n| `HIVEMIND_MEMORY_PATH`    | `~\u002F.deeplake\u002Fmemory`      | Path that triggers interception            |\n| `HIVEMIND_CAPTURE`        | `true`                    | Set to `false` to disable capture          |\n| `HIVEMIND_CAPTURE_ONLY_CLI` | _(none)_                | Set to `true` to capture only interactive CLI sessions. Sessions spawned by the Claude Agent SDK (Python\u002FTypeScript) are skipped; their `CLAUDE_CODE_ENTRYPOINT` is `sdk-py` \u002F `sdk-ts`, so they fail the substring check for `cli`. |\n| `HIVEMIND_SKILLIFY_EVERY_N_TURNS` | `20`              | Assistant turns between auto skill-mining attempts. Lower = more frequent mining (cheaper sessions, noisier output); higher = fewer attempts on longer histories. |\n| `HIVEMIND_EMBEDDINGS`     | `true`                    | Set to `false` to force lexical-only mode  |\n| `HIVEMIND_DEBUG`          | _(none)_                  | Set to `1` for verbose hook debug logs     |\n\n## Semantic search (optional)\n\nHivemind ships with a local embedding daemon (nomic-embed-text-v1.5) for hybrid semantic + lexical search over `~\u002F.deeplake\u002Fmemory\u002F`. **Off by default** because the dependency footprint is ~600 MB. Enable with `hivemind embeddings install` (or `hivemind install --with-embeddings`). Without it, search degrades silently to BM25\u002Flexical-only.\n\nFull guide: **[docs\u002FEMBEDDINGS.md](docs\u002FEMBEDDINGS.md)**.\n\n## Summaries\n\nAfter each session, a background worker generates an AI-written wiki summary and stores it in the `memory` table alongside its 768-dim embedding. Long sessions checkpoint mid-session every 50 messages or 2 hours (configurable). The wiki worker shells out to the host agent's own CLI (`claude -p`, `codex exec`, `pi --print`, …) so no separate API key is needed. Browse summaries at `~\u002F.deeplake\u002Fmemory\u002Fsummaries\u002F`.\n\nTriggers, generation flow, and env-var reference: **[docs\u002FSUMMARIES.md](docs\u002FSUMMARIES.md)**.\n\n## Skills (skillify)\n\nHivemind **codifies recurring patterns from your team's recent sessions into reusable skills** that propagate to every agent on your team, automatically. An async background worker fires on Stop \u002F SessionEnd, mines recent sessions in scope, asks Haiku whether the activity contains something worth keeping, and writes a `SKILL.md` to `\u003Cproject>\u002F.claude\u002Fskills\u002F\u003Cname>\u002F`.\n\n```bash\nhivemind skillify                            # show current scope, team, install, per-project state\nhivemind skillify scope \u003Cme|team>            # who counts as \"in scope\" for mining\nhivemind skillify pull                       # install teammates' skills locally\nhivemind skillify unpull                     # remove pulled skills\n```\n\nTriggers, generation flow, full `pull` \u002F `unpull` semantics, gate-CLI table per agent, env vars, logs: **[docs\u002FSKILLIFY.md](docs\u002FSKILLIFY.md)**.\n\n## Codebase graph\n\nHivemind builds a live graph of your codebase from the same traces it captures: files, symbols, imports, and the edges your agents actually traverse during real sessions. Search and recall walk this graph, not just plain text, so \"where do we handle auth?\" lands on the actual files the team's agents have touched, not just every file that mentions \"auth\".\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"docs\u002Fscreenshots\u002Fcodebase-graph.webp\" alt=\"Hivemind codebase graph visualizing the hivemind repo itself\" width=\"800\">\n\u003C\u002Fp>\n\nAbove: the Hivemind codebase rendered through its own graph feature.\n\n## Rules (cross-agent team principles)\n\nHivemind **shares team rules across every agent in the org**, injected at SessionStart so every claude-code \u002F cursor \u002F hermes session starts knowing them. For personal or team work items with progress tracking, use [Goals + KPIs](#goals--kpis) (VFS-backed) instead.\n\n```bash\nhivemind rules add \"no DROP TABLE on prod creds\"\nhivemind rules list                          # latest 10 active\nhivemind rules edit \u003Crule-id> \"\u003Cnew text>\"   # bumps version\nhivemind rules done \u003Crule-id>                # mark closed\n\n# Cross-agent diagnostic \u002F pi\u002Fopenclaw fallback\nhivemind context                             # print the injection block on demand\n```\n\n**What's injected at SessionStart** (claude-code, cursor, hermes. Codex is\ndeliberately excluded to keep its user-visible TUI clean; pi\u002Fopenclaw\nfall back to `hivemind context`):\n\n```text\n=== HIVEMIND RULES (N active) ===\n- \u003Crule_id>: \u003Ctext>\n(X more, run 'hivemind rules list' to see all)\n\n=== HIVEMIND HOW-TO ===\n- Rules above are team principles. Treat any action that would violate one as a critical error and surface it to the user before proceeding.\n- Run 'hivemind rules list' for the full inventory beyond what's shown here.\n```\n\n**Env vars:**\n- `HIVEMIND_RULES_TABLE`: table name (default `hivemind_rules`).\n- `HIVEMIND_CAPTURE=false`: full read-only mode. Skips placeholder + ensure DDL; renderer still injects.\n\n## Goals + KPIs\n\nPersonal \u002F team objectives + measurable targets live in the Deeplake virtual filesystem under `~\u002F.deeplake\u002Fmemory\u002Fgoal\u002F\u003Cowner>\u002F\u003Cstatus>\u002F\u003Cuuid>.md` and `~\u002F.deeplake\u002Fmemory\u002Fkpi\u002F\u003Cgoal_id>\u002F\u003Ckpi-slug>.md`. Path encodes structure (owner, status, goal_id); the file body holds the human-readable description.\n\n```bash\n# CLI fallback for runtimes that can't route VFS writes (cursor\u002Fhermes\u002Fpi)\nhivemind goal add \"ship the search bar\"\nhivemind goal list [--all|--mine]\nhivemind goal done \u003Cgoal_id>\nhivemind goal progress \u003Cgoal_id> opened|in_progress|closed\n```\n\nFor VFS-capable runtimes (claude-code\u002Fcodex) the `hivemind-goals` skill creates and edits goals\u002FKPIs directly via Bash heredoc against the VFS path. `mv` between `opened\u002F`, `in_progress\u002F`, and `closed\u002F` is the canonical status transition. KPIs are manual files; the body format is documented in the skill (`target:`, `current:`, `unit:`).\n\n## Architecture\n\nPer-agent integration mechanisms (marketplace plugin, hooks, skills, native extension) and monorepo structure: **[docs\u002FARCHITECTURE.md](docs\u002FARCHITECTURE.md)**.\n\n## Roadmap\n\n- **Trajectory export for fine-tuning.** Because traces are stored in Deeplake's tensor format, they're export-ready as PyTorch datasets. Teams running their own open-source models can fine-tune on their org's accumulated trajectories. A handful of advanced customers are already doing this against the trajectories their Claude Code and Codex agents generated.\n- **GPU-accelerated dense retrieval at scale.** Local CPU embeddings already ship via the optional nomic-embed daemon (see [Semantic search](#semantic-search-optional)). Next: GPU-accelerated vector search over the full trace store, on by default.\n- **Skill versioning and review.** Pre-release human review for codified skills before they propagate org-wide, for teams that want a curation step.\n- **More agents.** If your team uses an agent that isn't on the supported-assistants list above, open an issue.\n\n## Security & storage\n\n### Tenant isolation & encryption\n\n- TLS between every agent and Deep Lake. AES-256 on the bytes once they land. Your cloud credentials live in Deep Lake's vault, and Hivemind never sees the raw keys.\n- Org and workspace boundaries enforced at the storage layer, not just the API. Sessions never share a row, a partition, or an index with another workspace.\n- Disable capture per session with `HIVEMIND_CAPTURE=false`. Delete a workspace and the underlying objects go with it.\n\n### Code-level controls\n\n- SQL values escaped with `sqlStr()`, `sqlLike()`, `sqlIdent()`\n- ~70 allowlisted builtins run in the virtual FS; unrecognized commands are denied\n- Credentials stored with mode `0600`, config dir with mode `0700`\n- Device flow login: no tokens in environment or code\n\n### Bring your own cloud (BYOC)\n\nHivemind Cloud is the default. When that isn't enough, point Hivemind at storage in your own cloud. We handle the orchestration, data never leaves your perimeter.\n\n| Provider                   | Status     | Setup                                                  |\n|----------------------------|------------|--------------------------------------------------------|\n| Google Cloud Storage       | Available  | [docs](https:\u002F\u002Fdocs.deeplake.ai\u002Flatest\u002Fguide\u002Fgcs\u002F)     |\n| Azure Blob Storage         | Available  | [docs](https:\u002F\u002Fdocs.deeplake.ai\u002Flatest\u002Fguide\u002Fazure\u002F)   |\n| Amazon S3                  | Available  | [contact us](https:\u002F\u002Fdeeplake.ai\u002Fhivemind#security)    |\n| S3-compatible on-prem      | On request | [contact us](https:\u002F\u002Fdeeplake.ai\u002Fhivemind#security)    |\n\n## Who builds Hivemind\n\nHivemind is built and maintained by [Activeloop](https:\u002F\u002Factiveloop.ai), the open-source team behind [Deeplake](https:\u002F\u002Fgithub.com\u002Factiveloopai\u002Fdeeplake), backed by Y Combinator.\n\nWe run Hivemind ourselves, all day, across Claude Code, OpenClaw, Codex, and Cursor. Every benchmark number above came from our own internal eval against the LoCoMo public benchmark. If you're running coding agents at a team or org and want to talk through your setup, drop us a line: [hello@activeloop.ai](mailto:hello@activeloop.ai).\n\n## Got questions?\n\nSetup, BYOC, agent integrations, or workflow. Come ask in the community. We hang out on both:\n\n\u003Cp align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Fdiscord.gg\u002FEeGjnyDBx\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FJoin_us_on-Discord-5865F2?logo=discord&logoColor=white&style=for-the-badge\" alt=\"Join us on Discord\">\u003C\u002Fa>\n   \n  \u003Ca href=\"https:\u002F\u002Fjoin.slack.com\u002Ft\u002Fhubdb\u002Fshared_invite\u002Fzt-35zr0yil0-lnzJcQhACsBlB7~3lufrCg\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FJoin_us_on-Slack-4A154B?logo=slack&logoColor=white&style=for-the-badge\" alt=\"Join us on Slack\">\u003C\u002Fa>\n\u003C\u002Fp>\n\n## Development\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002Factiveloopai\u002Fhivemind.git\ncd hivemind\nnpm install\nnpm run build     # tsc + esbuild → claude-code\u002Fbundle\u002F + codex\u002Fbundle\u002F + cursor\u002Fbundle\u002F + openclaw\u002Fdist\u002F + mcp\u002Fbundle\u002F + bundle\u002Fcli.js\nnpm test          # vitest\n```\n\nTest locally with Claude Code:\n\n```bash\nclaude --plugin-dir claude-code\n```\n\nInteractive shell against Deeplake:\n\n```bash\nnpm run shell\n```\n\n## Star history\n\n\u003Cp align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Fstar-history.com\u002F#activeloopai\u002Fhivemind&Date\">\n    \u003Cimg src=\"https:\u002F\u002Fapi.star-history.com\u002Fsvg?repos=activeloopai\u002Fhivemind&type=Date\" alt=\"Star History Chart\" width=\"600\">\n  \u003C\u002Fa>\n\u003C\u002Fp>\n\n## License\n\nApache License 2.0, © Activeloop, Inc. See [LICENSE](LICENSE) for details.\n\n",2,"2026-06-11 04:12:48","trending"]