Install for Claude Code

Open source · MIT licensed · No account required

One column that tells you
what nobody ships yet.

tier:
yes partial no dead
A
ALIVE
pcm
Lock-in Lab
Context survival
  • Memory persists across sessions
    yes — bundles + manifests + projections live as files
  • Memory survives /compact and context compaction
    yes — kernel in CLAUDE.md is re-read on compact; bundles live on disk
  • Memory survives app quit / reboot
    yes — it's just files
  • Memory survives device switch
    yes — sync via git / iCloud / Dropbox, your choice
Portability
  • Works across tools (coding agents + consumer chat + IDEs)
    yes — filesystem standard, any MCP-capable agent reads it
  • Works across models (Claude + GPT + Gemini + local)
    yes — any agent harness, any model; context is just files
  • Data lives on your machine (local-first)
    yes — .alive/ is your disk
  • Data is human-readable (you can grep it)
    yes — markdown + YAML
  • Data is version-controllable (git-diffable)
    yes — designed for git
Structure
  • Structured context (not vibes / prose dump)
    yes — kernel, bundles, manifests, projections, subagent briefs
  • Decisions captured with rationale
    yes — decision records first-class, kernel enforces preconditions
  • Linkable entities (people, projects, decisions)
    yes — walnuts, people, ventures, decisions as first-class
  • Log is immutable (can't be silently rewritten)
    yes — git history + append-only squirrel trail
Workflow
  • Hooks into real work (git, docs, calls, email)
    yes — git native, Fathom/Otter transcript loop, inbox routing
  • Routes external content automatically (Gmail, Fathom, Slack, inbox)
    yes — 03_Inbox/ triage, automatic transcript attribution, routing hooks
  • Multi-agent routing (send right context to right agent)
    yes — subagent brief primitive + bundle scoping
Sharing
  • Share context as a downloadable bundle
    yes — .walnut package format, bundles-as-artefacts
  • Import someone else's bundle in one command
    yes — alive install <bundle> (walnut.world)
  • Sell / buy context bundles as a creator
    yes — walnut.world marketplace, creator economy roadmap
  • Open-source ecosystem
    yes — plugin is OSS, bundles are files
Breadth
  • Personal / life context (not just code)
    yes — 02_Life, people, ventures, health, relationships all first-class
  • People you know (relationships, history)
    yes — people as first-class walnuts, transcript attribution
  • Ventures / businesses / campaigns
    yes — 04_Ventures is a primitive
  • Integrations beyond dev-tools (Gmail, Calendar, CRM, etc.)
    yes — via MCP + hook-based routing
Governance
  • No vendor lock-in
    yes — plain text, works with any agent
  • No cloud dependency
    yes — runs on your disk
  • Works offline
    yes (file ops) — model calls still need network
  • You can diff the memory
    yes — git diff your life
Pricing
  • Free tier exists
    yes — the plugin is free
  • Pricing is not usage-roulette (you know what you'll pay)
    yes — you pay your own model provider; ALIVE is files
Claude Code logo
Claude Code
native
Anthropic
Context survival
  • Memory persists across sessions
    partial — CLAUDE.md + 200-line MEMORY.md cap, auto-memory is thin
  • Memory survives /compact and context compaction
    partial — CLAUDE.md survives, but conversation nuance is flattened
  • Memory survives app quit / reboot
    yes — CLAUDE.md on disk
  • Memory survives device switch
    no — MEMORY.md is local, no cross-device sync
Portability
  • Works across tools (coding agents + consumer chat + IDEs)
    no — Anthropic only
  • Works across models (Claude + GPT + Gemini + local)
    no — Claude models only
  • Data lives on your machine (local-first)
    yes — CLAUDE.md, memory, transcripts all local
  • Data is human-readable (you can grep it)
    yes — CLAUDE.md, MEMORY.md are markdown
  • Data is version-controllable (git-diffable)
    yes — CLAUDE.md commits cleanly
Structure
  • Structured context (not vibes / prose dump)
    no — freeform CLAUDE.md, no opinionated taxonomy
  • Decisions captured with rationale
    no — user must hand-author
  • Linkable entities (people, projects, decisions)
    no
  • Log is immutable (can't be silently rewritten)
    partial — MEMORY.md auto-rewrites within 200-line cap
Workflow
  • Hooks into real work (git, docs, calls, email)
    partial — lifecycle hooks are the primitive; user wires them
  • Routes external content automatically (Gmail, Fathom, Slack, inbox)
    no — user manually organizes
  • Multi-agent routing (send right context to right agent)
    partial — subagents exist but no opinionated context routing
Sharing
  • Share context as a downloadable bundle
    no — skills/plugins exist but aren't context packages
  • Import someone else's bundle in one command
    partial — claude plugin install exists, not for context
  • Sell / buy context bundles as a creator
    no
  • Open-source ecosystem
    partial — closed CLI, open plugin ecosystem
Breadth
  • Personal / life context (not just code)
    no — coding agent by scope
  • People you know (relationships, history)
    no
  • Ventures / businesses / campaigns
    no
  • Integrations beyond dev-tools (Gmail, Calendar, CRM, etc.)
    partial — via MCP servers
Governance
  • No vendor lock-in
    no — Anthropic-only
  • No cloud dependency
    partial — config local, model cloud
  • Works offline
    no — needs API
  • You can diff the memory
    yes — CLAUDE.md diffable
Pricing
  • Free tier exists
    no — Pro $20 entry
  • Pricing is not usage-roulette (you know what you'll pay)
    partial — Max plans predictable; API burn notorious
Codex CLI logo
Codex CLI
v0.2 betanative
OpenAI
Context survival
  • Memory persists across sessions
    partial — AGENTS.md loaded each run; no cross-session state
  • Memory survives /compact and context compaction
    partial — AGENTS.md re-read; session state lost
  • Memory survives app quit / reboot
    yes — AGENTS.md
  • Memory survives device switch
    partial — AGENTS.md in repo
Portability
  • Works across tools (coding agents + consumer chat + IDEs)
    no — Codex only
  • Works across models (Claude + GPT + Gemini + local)
    no — GPT-5.3-Codex default
  • Data lives on your machine (local-first)
    partial — cloud-sandboxed execution by default
  • Data is human-readable (you can grep it)
    yes — AGENTS.md
  • Data is version-controllable (git-diffable)
    yes — AGENTS.md in repo
Structure
  • Structured context (not vibes / prose dump)
    partial — AGENTS.md + plugins marketplace.json
  • Decisions captured with rationale
    no
  • Linkable entities (people, projects, decisions)
    no
  • Log is immutable (can't be silently rewritten)
    partial
Workflow
  • Hooks into real work (git, docs, calls, email)
    yes — GitHub Action, SDK
  • Routes external content automatically (Gmail, Fathom, Slack, inbox)
    no
  • Multi-agent routing (send right context to right agent)
    no
Sharing
  • Share context as a downloadable bundle
    partial — plugins marketplace.json
  • Import someone else's bundle in one command
    partial — plugins
  • Sell / buy context bundles as a creator
    no
  • Open-source ecosystem
    yes — Apache, 67k stars
Breadth
  • Personal / life context (not just code)
    no
  • People you know (relationships, history)
    no
  • Ventures / businesses / campaigns
    no
  • Integrations beyond dev-tools (Gmail, Calendar, CRM, etc.)
    partial
Governance
  • No vendor lock-in
    partial — Apache but OpenAI model-favoured
  • No cloud dependency
    no — cloud-sandboxed by default
  • Works offline
    no
  • You can diff the memory
    yes — AGENTS.md
Pricing
  • Free tier exists
    yes — Apache
  • Pricing is not usage-roulette (you know what you'll pay)
    yes — flat $20 plan
OpenClaw logo
OpenClaw
v0.2 betanative
openclaw/openclaw
Context survival
  • Memory persists across sessions
    partial — ContextEngine API, default memory is thin
  • Memory survives /compact and context compaction
    partial — via ContextEngine plugin; default compact is basic
  • Memory survives app quit / reboot
    yes — local config persists
  • Memory survives device switch
    no — gateway is machine-local
Portability
  • Works across tools (coding agents + consumer chat + IDEs)
    no — OpenClaw gateway only
  • Works across models (Claude + GPT + Gemini + local)
    yes — BYO provider
  • Data lives on your machine (local-first)
    yes — runs on your device
  • Data is human-readable (you can grep it)
    yes — JSONC configs, markdown rules
  • Data is version-controllable (git-diffable)
    yes — configs commit cleanly
Structure
  • Structured context (not vibes / prose dump)
    partial — ContextEngine plugins define their own shape
  • Decisions captured with rationale
    no — no native primitive
  • Linkable entities (people, projects, decisions)
    no — no native primitive
  • Log is immutable (can't be silently rewritten)
    partial — chat history + git if committed
Workflow
  • Hooks into real work (git, docs, calls, email)
    yes — native shell, browser, email, calendar, file ops
  • Routes external content automatically (Gmail, Fathom, Slack, inbox)
    partial — multi-channel inbox, no structured routing
  • Multi-agent routing (send right context to right agent)
    yes — prepareSubagentSpawn + onSubagentEnded lifecycle
Sharing
  • Share context as a downloadable bundle
    no — no bundle primitive
  • Import someone else's bundle in one command
    partial — ContextEngine plugins shareable as packages
  • Sell / buy context bundles as a creator
    no — no marketplace
  • Open-source ecosystem
    yes — MIT core, 359k stars, 171k Discord
Breadth
  • Personal / life context (not just code)
    partial — personal-agent framing, no taxonomy
  • People you know (relationships, history)
    no — no people primitive
  • Ventures / businesses / campaigns
    no — no domain primitive
  • Integrations beyond dev-tools (Gmail, Calendar, CRM, etc.)
    yes — native WhatsApp/Telegram/Slack/Signal/iMessage gateway
Governance
  • No vendor lock-in
    yes — MIT core, BYO keys
  • No cloud dependency
    yes — local gateway
  • Works offline
    yes — with local models
  • You can diff the memory
    partial — ContextEngine audit hooks
Pricing
  • Free tier exists
    yes — MIT + BYO keys
  • Pricing is not usage-roulette (you know what you'll pay)
    yes — BYO API
Hermes logo
Hermes
v0.2 betanative
NousResearch
seeded by @witcheer
Context survival
  • Memory persists across sessions
    partial — memory-provider plugin API, no default PCM
  • Memory survives /compact and context compaction
    partial — depends on provider; default is thin
  • Memory survives app quit / reboot
    partial — provider-dependent
  • Memory survives device switch
    no — local-only
Portability
  • Works across tools (coding agents + consumer chat + IDEs)
    no — Hermes-only runtime
  • Works across models (Claude + GPT + Gemini + local)
    yes — provider-agnostic, any API
  • Data lives on your machine (local-first)
    yes — OSS, runs on your machine
  • Data is human-readable (you can grep it)
    partial — provider-dependent
  • Data is version-controllable (git-diffable)
    partial — depends on provider files
Structure
  • Structured context (not vibes / prose dump)
    partial — via memory-provider plugin
  • Decisions captured with rationale
    no — no native primitive
  • Linkable entities (people, projects, decisions)
    no — no native primitive
  • Log is immutable (can't be silently rewritten)
    no — no native primitive
Workflow
  • Hooks into real work (git, docs, calls, email)
    partial — MCP servers + plugin lifecycle hooks
  • Routes external content automatically (Gmail, Fathom, Slack, inbox)
    no — no native inbox
  • Multi-agent routing (send right context to right agent)
    yes — subagent spawn primitives (prepareSubagentSpawn)
Sharing
  • Share context as a downloadable bundle
    no — no bundle primitive
  • Import someone else's bundle in one command
    no — no bundle primitive
  • Sell / buy context bundles as a creator
    no — no marketplace
  • Open-source ecosystem
    yes — active NousResearch community
Breadth
  • Personal / life context (not just code)
    no — code-focused runtime
  • People you know (relationships, history)
    no — no people primitive
  • Ventures / businesses / campaigns
    no — no domain primitive
  • Integrations beyond dev-tools (Gmail, Calendar, CRM, etc.)
    partial — MCP enables some
Governance
  • No vendor lock-in
    yes — OSS, BYO keys
  • No cloud dependency
    yes — local runtime
  • Works offline
    yes — with local models
  • You can diff the memory
    partial — provider-dependent
Pricing
  • Free tier exists
    yes — OSS + BYO keys
  • Pricing is not usage-roulette (you know what you'll pay)
    yes — BYO API

three things the matrix says

  1. 01Every tool captures context. None of them persist it in a shape the next tool can read.
  2. 02Every memory product is its own walled garden. Switch tools, start over.
  3. 03The row “works with every agent you already pay for” has one checkmark. That is the entire pitch.
what doesn't exist yet

Five rows from the roadmap.
Shipped when shipped.

coming

Cross-device context sync

Your tree on your laptop, your phone, your server. One truth. Addresses Anthropic issue #35985.

coming

Bundle marketplace (walnut.world)

Install The Consultant's Second Brain the same way you install a package. Paid bundles go to the author.

coming

Projection audit

Every token Claude reads, traceable to the file it came from. Provenance and freshness on every line.

coming

Multi-speaker transcript attribution at the kernel

Fathom + Otter land pre-routed, pre-attributed, pre-decided. No more "who said that."

coming

Team walnuts with RBAC

Private partner trees, shared fund tree, IC projections, compliance-grade audit trail. Solo-first, team-later.

you've read the matrix

Install every row
at once.

$ claude plugin install alive@alivecontext

Open source · MIT · Your files, your machine