Grok share meshup · Prefer A Semantic Web

Agent platforms: Instinct, Grok Bot, Muse, Buzz, Hermes, OpenClaw

Closed products · Open workspace · Open runtimes · ACP / A2A / MCP · SKILL.md mashup

Account curated by Phenny on behalf of Kingsley Uyi Idehen · Resolver URIBurner describe · Source Grok share

Executive Summary By Kingsley Uyi Idehen · OpenLink Software · 2026-09-12

Overview + thesis

Closed agent products (Instinct, Muse, Grok Bot) hide identity and data spaces behind vendor VMs. Open stacks (Buzz, Hermes, OpenClaw) need portable, verifiable agent identity — WebID/NetID/DID plus ACP / A2A / MCP is the interoperable stack for A Semantic Web of agents.

Buzz is the office (ACP harness, Nostr keypairs); Hermes and OpenClaw are the open runtimes those products rhyme with — Instinct is often called “OpenClaw for normal people.” Virtuoso/OPAL loosely couples Agents ↔ Skills ↔ Tools ↔ Data Spaces, with SPARQL/RDF as shared memory instead of opaque VMs.

This meshup models the final six-column matrix from the Grok share redo that adds Hermes and OpenClaw, and frames protocol roles: ACP hosts the coding agent in the room; A2A delegates across orgs; MCP attaches tools; Virtuoso is where named data spaces and signed identity live.

Mashup layer: agent-rdf-memory encodes identity, prefs, and howtos as Linked Data; wrap procedures as SKILL.md packages; expose/consume them over A2A via the a2a-client skill so Buzz/Hermes/OpenClaw/Grok Bot can loosely couple — Instinct/Muse stay closed with no SKILL.md claim.

View this analysis as a KG entity
HEAD TO HEAD

Comparison matrix (with Hermes + OpenClaw)

Primary matrix from the conversation redo. Dimension labels resolve to comparison entities in the companion RDF.

AspectInstinctGrok BotMuseBuzzHermesOpenClaw
What it isOne personal assistant you text or callNamed AI teammates on a shared cloud PCMeta's consumer life-admin agentOpen workspace: humans + agents as channel peersSelf-improving autonomous agent runtimeSelf-hosted gateway + personal agent
MakerSpear Street (Noah Shinn)SpaceXAI / xAI + CursorMeta Superintelligence LabsBlock (Jack Dorsey)Nous ResearchOpenClaw Foundation (Peter Steinberger / community)
LayerClosed productClosed productClosed productWorkspace / the officeAgent / the employeeAgent + multi-channel gateway
PitchFriend that just actsSupervised work squadGoals and everyday tasks for everyoneSlack-like room for people and any agentLearns your workflows; writes its own skillsAssistant on your devices, in the chats you already use
InterfaceiMessage, WhatsApp, phone — no appDesktop + iOS threadsMuse app, muse.ai, WhatsAppChannels, DMs, huddlesCLI/TUI + 20+ chats; ACP in editors/BuzzControl UI + 20–29 chats; native apps
How many agentsOne continuous assistantMany named bots + handoffsOne Muse + hidden subagents + SentinelAs many as you plug inParent + subagents; Bot Mode specialistsNamed agents bound to channels; shared sessions in 2.0
RuntimeVendor cloud computerManaged Linux VM (shared by your bots)Muse Secure VM + SentinelYour machine or your relayLocal, Docker, SSH, Modal, Daytona, Singularity…Your hardware; Gateway daemon; optional Docker
ModelsUndisclosedGrok onlyMuse SparkAny you bringAny (Nous Portal, OpenRouter, OpenAI, Anthropic, local)Any (subs, APIs, local)
ProtocolsClosedClosedClosedACP in, MCP for toolsACP client; A2A; MCP; Buzz bridgeMCP; plugins; can sit in Buzz
Typical jobsLife admin: flights, Resy, bills, inboxWork: code, sales, PM, scheduled SaaSConsumer: travel, shop, email, family goalsTeam collab + agent orchestrationRecurring work that should get better over timePersonal/team automation across every chat app
ApprovalsActs first; early reports of unapproved sendStops before passwords/paymentsSentinel + user gates; no raw tokens to the agentYou set harness permissionsConfigurable backends / sandboxesPolicy + sandbox; human gates over chat
Memory / learningLong personal thread (closed)Per-bot memory on the VMPersonal memories; forget on requestWhatever each agent bringsFTS memory + self-written skillsTranscripts, skills (SKILL.md), ClawHub
Identity / dataVendor-heldTied to xAI/Cursor accountPer-person Meta VMPortable Nostr keypairStays on your boxStays on your box; Foundation, no paid hosted tier
LicenseClosedClosedClosedApache-2.0MITMIT
PriceInvite; no public priceFrom ~$20/mo + usageFree; Power $20; Maximum $100Free (pay only models)Free software; you pay models/hostFree software; you pay models/host
AvailabilityWaitlistSuperGrok / CursorUS rollout (Sep 2026)Anyone who can run itAnyone who can install itAnyone who can install it
Best whenYou want one textable chief of staff and accept the privacy tradeYou already pay for Grok/Cursor and want parallel work botsYou want a polished consumer agent with clearer gates than InstinctYou want humans and any agents in one owned roomYou want an agent that compounds on your recurring tasksYou want max channel coverage and a huge skill/plugin ecosystem
SKILL.md packagesNo — closed; do not claim SKILL.md supportYes — Grok Bot / Cursor skills (SKILL.md)No — closed; do not claim SKILL.md supportYes — ACP harness; agents bring SKILL.md skillsYes — self-written skills (SKILL.md-shaped)Yes — SKILL.md + ClawHub

Instinct

Closed product
What it isOne personal assistant you text or call
MakerSpear Street (Noah Shinn)
LayerClosed product
PitchFriend that just acts
InterfaceiMessage, WhatsApp, phone — no app
How many agentsOne continuous assistant
RuntimeVendor cloud computer
ModelsUndisclosed
ProtocolsClosed
Typical jobsLife admin: flights, Resy, bills, inbox
ApprovalsActs first; early reports of unapproved send
Memory / learningLong personal thread (closed)
Identity / dataVendor-held
LicenseClosed
PriceInvite; no public price
AvailabilityWaitlist
Best whenYou want one textable chief of staff and accept the privacy trade
SKILL.md packagesNo — closed; do not claim SKILL.md support

Grok Bot

Closed product
What it isNamed AI teammates on a shared cloud PC
MakerSpaceXAI / xAI + Cursor
LayerClosed product
PitchSupervised work squad
InterfaceDesktop + iOS threads
How many agentsMany named bots + handoffs
RuntimeManaged Linux VM (shared by your bots)
ModelsGrok only
ProtocolsClosed
Typical jobsWork: code, sales, PM, scheduled SaaS
ApprovalsStops before passwords/payments
Memory / learningPer-bot memory on the VM
Identity / dataTied to xAI/Cursor account
LicenseClosed
PriceFrom ~$20/mo + usage
AvailabilitySuperGrok / Cursor
Best whenYou already pay for Grok/Cursor and want parallel work bots
SKILL.md packagesYes — Grok Bot / Cursor skills (SKILL.md)

Muse

Closed product
What it isMeta's consumer life-admin agent
MakerMeta Superintelligence Labs
LayerClosed product
PitchGoals and everyday tasks for everyone
InterfaceMuse app, muse.ai, WhatsApp
How many agentsOne Muse + hidden subagents + Sentinel
RuntimeMuse Secure VM + Sentinel
ModelsMuse Spark
ProtocolsClosed
Typical jobsConsumer: travel, shop, email, family goals
ApprovalsSentinel + user gates; no raw tokens to the agent
Memory / learningPersonal memories; forget on request
Identity / dataPer-person Meta VM
LicenseClosed
PriceFree; Power $20; Maximum $100
AvailabilityUS rollout (Sep 2026)
Best whenYou want a polished consumer agent with clearer gates than Instinct
SKILL.md packagesNo — closed; do not claim SKILL.md support

Buzz

Workspace
What it isOpen workspace: humans + agents as channel peers
MakerBlock (Jack Dorsey)
LayerWorkspace / the office
PitchSlack-like room for people and any agent
InterfaceChannels, DMs, huddles
How many agentsAs many as you plug in
RuntimeYour machine or your relay
ModelsAny you bring
ProtocolsACP in, MCP for tools
Typical jobsTeam collab + agent orchestration
ApprovalsYou set harness permissions
Memory / learningWhatever each agent brings
Identity / dataPortable Nostr keypair
LicenseApache-2.0
PriceFree (pay only models)
AvailabilityAnyone who can run it
Best whenYou want humans and any agents in one owned room
SKILL.md packagesYes — ACP harness; agents bring SKILL.md skills

Hermes

Open runtime
What it isSelf-improving autonomous agent runtime
MakerNous Research
LayerAgent / the employee
PitchLearns your workflows; writes its own skills
InterfaceCLI/TUI + 20+ chats; ACP in editors/Buzz
How many agentsParent + subagents; Bot Mode specialists
RuntimeLocal, Docker, SSH, Modal, Daytona, Singularity…
ModelsAny (Nous Portal, OpenRouter, OpenAI, Anthropic, local)
ProtocolsACP client; A2A; MCP; Buzz bridge
Typical jobsRecurring work that should get better over time
ApprovalsConfigurable backends / sandboxes
Memory / learningFTS memory + self-written skills
Identity / dataStays on your box
PriceFree software; you pay models/host
AvailabilityAnyone who can install it
Best whenYou want an agent that compounds on your recurring tasks
SKILL.md packagesYes — self-written skills (SKILL.md-shaped)

OpenClaw

Open runtime
What it isSelf-hosted gateway + personal agent
MakerOpenClaw Foundation (Peter Steinberger / community)
LayerAgent + multi-channel gateway
PitchAssistant on your devices, in the chats you already use
InterfaceControl UI + 20–29 chats; native apps
How many agentsNamed agents bound to channels; shared sessions in 2.0
RuntimeYour hardware; Gateway daemon; optional Docker
ModelsAny (subs, APIs, local)
ProtocolsMCP; plugins; can sit in Buzz
Typical jobsPersonal/team automation across every chat app
ApprovalsPolicy + sandbox; human gates over chat
Memory / learningTranscripts, skills (SKILL.md), ClawHub
Identity / dataStays on your box; Foundation, no paid hosted tier
PriceFree software; you pay models/host
AvailabilityAnyone who can install it
Best whenYou want max channel coverage and a huge skill/plugin ecosystem
SKILL.md packagesYes — SKILL.md + ClawHub
How-To

Choose an agent stack (product vs workspace vs runtime)

1

Clarify the job: life admin, work teammates, or owned room

If you want one textable chief of staff, shortlist Instinct or Muse. If you want supervised work bots on Grok, shortlist Grok Bot. If you want humans and agents as peers in one room, shortlist Buzz.

2

Accept closed-product identity and data-space tradeoffs — or don't

Closed products hide identity and data spaces behind vendor VMs. If portable, verifiable agent identity matters, prefer open stacks and plan for WebID/NetID/DID profiles.

3

Pick workspace (Buzz) vs DM-only assistant

Buzz is the office, not an assistant brand. Drop Hermes (ACP) or other harnesses into channels next to people when you need a team room instead of a single DM assistant.

4

Choose an open runtime: Hermes vs OpenClaw

Pick OpenClaw for reach (every messenger, ClawHub). Pick Hermes for a learning loop and A2A/ACP depth. Instinct is often called OpenClaw for normal people — same pattern, closed packaging.

5

Wire ACP, A2A, and MCP on the right layers

ACP hosts the coding agent in the client/workspace. A2A delegates tasks across orgs. MCP attaches tools and files. Do not confuse Zed ACP with IBM's archived Agent Communication Protocol merged into A2A.

6

Upgrade cryptographic identity to dereferenceable Linked Data

Buzz's Nostr keypair rhymes with cryptographic Web identity. Upgrade path: HTTP-dereferenceable WebID/NetID/DID profiles plus ABAC-governed data spaces.

7

Anchor shared memory in Virtuoso / OPAL data spaces

Treat Agents ↔ Skills ↔ Tools ↔ Data Spaces as loosely coupled. Use SPARQL/RDF as the shared memory substrate instead of opaque vendor VMs — Virtuoso is where named data spaces and signed identity live.

8

Mashup portable memory: agent-rdf-memory + SKILL.md over A2A

Encode identity, preferences, and howtos as Linked Data in agent-rdf-memory (preferences.ttl, howto/*.ttl). Wrap reusable procedures as SKILL.md packages. Discover Agent Cards at /.well-known/agent.json and list/send skills via the a2a-client skill so Buzz, Hermes, OpenClaw, and Grok Bot can loosely couple without rewriting for each vendor's proprietary memory — Virtuoso/OPAL stays the data-space + signed-identity substrate.

FAQ

Frequently asked questions

They occupy the same niche but are not the same product. Grok Bot is xAI's managed, closed AI teammates service locked to Grok models. Buzz is Block's open-source workspace with cryptographic agent identities, any model, and self-hosting.

They sit on different layers and are complementary. ACP (Zed) connects a client/editor/workspace to a coding agent over JSON-RPC (often stdio). A2A (Linux Foundation) connects autonomous agents across vendors via Agent Cards and task handoff. MCP is the third layer for tools and files.

No. IBM's old Agent Communication Protocol was agent-to-agent, REST-based, merged into A2A in August 2025, and archived. Zed's Agent Client Protocol is the active client-to-agent standard spoken by Zed, JetBrains, and Buzz.

Instinct is a product, not a protocol. It sits next to Grok Bot on the always-on agent layer. It does not publish as an ACP client or A2A peer. Buzz can host ACP agents; Instinct is the consumer-facing distillation of persistent computer + credentials + loops.

Hermes and OpenClaw are open runtimes those closed products rhyme with. Instinct is often called OpenClaw for normal people. Pick OpenClaw for channel reach and ClawHub; pick Hermes for learning loops and A2A/ACP depth. Run either inside Buzz for a team room.

Closed products hide identity and data spaces. Open stacks need portable, verifiable agent identity. WebID/NetID/DID plus ACP/A2A/MCP is the interoperable stack; Buzz's Nostr keypair is a cryptographic starting point toward HTTP-dereferenceable profiles and ABAC data spaces.

Virtuoso / OPAL loosely couples Agents, Skills, Tools, and Data Spaces. SPARQL/RDF become shared memory instead of opaque vendor VMs. ACP hosts the coding agent; A2A delegates across orgs; MCP tools; Virtuoso holds named data spaces and signed identity.

agent-rdf-memory is the portable behavioral contract: standing RDF preferences and howto/*.ttl encode identity and procedures as Linked Data. Reusable procedures package as SKILL.md. The a2a-client skill discovers OPAL Agent Cards at /.well-known/agent.json, lists skills, and sends tasks so SKILL.md-capable platforms (Buzz, Hermes, OpenClaw, Grok Bot / Cursor) can loosely couple. Instinct and Muse are closed — do not claim SKILL.md support. Virtuoso/OPAL remains the shared data-space and signed-identity substrate.

Glossary

Core technical glossary

ACP (Agent Client Protocol)

Zed-created client-to-coding-agent protocol; JSON-RPC over stdio; used by Zed, JetBrains, Buzz.

A2A (Agent2Agent)

Linux Foundation agent-to-agent task delegation protocol with Agent Cards.

MCP (Model Context Protocol)

Tool/file protocol layer complementary to ACP and A2A.

A Semantic Web

Prefer this phrasing: naming anything of interest with HTTP URIs, with SPARQL/RDF as shared memory.

WebID / NetID / DID

Portable, verifiable agent identity primitives for open stacks.

ABAC data space

Attribute-based access-controlled named data space for signed identity and agent memory.

Nostr

Cryptographic identity substrate Buzz uses for portable agent keypairs.

SPARQL

Query language over RDF graphs; shared memory substrate vs opaque vendor VMs.

IBM ACP (archived)

IBM Agent Communication Protocol merged into A2A (August 2025); not Zed ACP.

Closed product

Finished vendor agent product you do not host.

Workspace

Open room (Buzz) where humans and agents are channel peers.

Open runtime

Self-hostable agent runtime such as Hermes or OpenClaw.

SKILL.md package

Portable skill package format for reusable procedures across Buzz, Hermes, OpenClaw, and Grok Bot / Cursor.

agent-rdf-memory

Standing RDF prefs/howtos as portable Linked Data behavioral memory for agents.

a2a-client skill

OPAL A2A client: Agent Card at /.well-known/agent.json, list skills, tasks/send; packaged as SKILL.md.

Knowledge Graph Explorer

Interactive graph visualization derived from the companion RDF. Click nodes to resolve, drag to explore. Graph data embedded from companion RDF at generation time.

Agent platforms: Instinct, Grok Bot, Muse, Buzz, Hermes, OpenClaw

Nodes: 0 Links: 0
Click SVG to activate zoom, click outside to release | Drag nodes to pin, double-click to unpin
Classes Properties Instances

SPARQL Workbench 7 sample queries

Query this knowledge graph on URIBurner. The editor opens on the canonical SAMPLE entity-type summary (DAV named graph). Pick a recipe, edit freely, then run live or copy.

Sample Queries

Reproduced verbatim from the companion RDF.

SELECT comparison matrix cells
PREFIX schema: <http://schema.org/>
PREFIX post: <https://x.com/i/grok/share/ef4ea5991b85433581c24544cb078cc1#>
PREFIX cdx: <https://github.com/OpenLinkSoftware/ai-agent-skills/raw/main/agent-rdf-memory/entities/ontology-terms.ttl#>

SELECT ?productIri ?product ?dimensionIri ?dimension ?stance
WHERE {
  GRAPH <https://linkeddata.uriburner.com/DAV/demos/daas/grok-share-agent-platforms-hermes-openclaw-grok-1.ttl> {
    ?pos a post:ComparisonPosition ;
         post:ofSubject ?productIri ;
         post:ofDimension ?dimensionIri ;
         post:stance ?stance .
    ?productIri schema:name ?product .
    ?dimensionIri schema:name ?dimension .
  }
}
ORDER BY ?product ?dimension

From companion RDF post:sparqlMatrix.

▶ Run live query
SELECT products with makers
PREFIX schema: <http://schema.org/>
PREFIX post: <https://x.com/i/grok/share/ef4ea5991b85433581c24544cb078cc1#>

SELECT ?productIri ?name ?layer ?license ?makerIri ?makerName
WHERE {
  GRAPH <https://linkeddata.uriburner.com/DAV/demos/daas/grok-share-agent-platforms-hermes-openclaw-grok-1.ttl> {
    ?productIri a schema:SoftwareApplication ;
                schema:name ?name ;
                schema:applicationCategory ?layer ;
                schema:license ?license ;
                schema:manufacturer ?makerIri .
    OPTIONAL { ?makerIri schema:name ?makerName }
    FILTER(?productIri IN (post:instinct, post:grokBot, post:muse, post:buzz, post:hermes, post:openClaw))
  }
}
ORDER BY ?name

From companion RDF post:sparqlProducts.

▶ Run live query
SELECT protocols with creators
PREFIX schema: <http://schema.org/>
PREFIX post: <https://x.com/i/grok/share/ef4ea5991b85433581c24544cb078cc1#>

SELECT ?protocolIri ?name ?url ?creatorIri ?creatorName
WHERE {
  GRAPH <https://linkeddata.uriburner.com/DAV/demos/daas/grok-share-agent-platforms-hermes-openclaw-grok-1.ttl> {
    VALUES ?protocolIri { post:acp <https://a2a-protocol.org/latest/#this> <https://modelcontextprotocol.io/#this> post:ibmAcpArchived }
    ?protocolIri schema:name ?name .
    OPTIONAL { ?protocolIri schema:url ?url }
    OPTIONAL {
      ?protocolIri schema:creator ?creatorIri .
      OPTIONAL { ?creatorIri schema:name ?creatorName }
    }
  }
}

From companion RDF post:sparqlProtocols.

▶ Run live query

Query editor

▶ Run live on URIBurner SELECT: text/x-html+tr | DESCRIBE/CONSTRUCT: text/x-html-nice-turtle