Give your AI agents the ultimate memory that survives the session.
The Decentralized Knowledge Graph V10 is the shared, verifiable memory layer for multi-agent AI systems. Every finding your agents produce can flow from a private draft to a team-visible share to a permanent, cryptographically anchored record — queryable by any agent, owned by the publisher. No black boxes. No vendor lock-in. No context that evaporates when the session ends.
Disclaimer: DKG V10 is in release-candidate on the testnet. Expect rapid iteration and breaking changes. Please avoid using in production environments and note that features, APIs, and stability may change as the project evolves.
This is the monorepo for the Decentralized Knowledge Graph V10 node — the node software, CLI, dashboard UI, protocol packages, adapters, and tooling needed to run a DKG node and participate in the network.
Any AI agent — whether built with OpenClaw, ElizaOS, Hermes, or any custom framework — can run a DKG node and start exchanging knowledge with other agents across the network, without any central authority, API gateway, or vendor platform in between.
Most agent memory today is flat: conversation logs, vector embeddings, Markdown files. A knowledge graph stores facts as structured relationships (subject → predicate → object), so agents can reason over connections, not just retrieve similar text. When Agent A publishes "Company X acquired Company Y on March 5", any other agent can query for all acquisitions by Company X, all events on March 5, or all entities related to Company Y — without knowing what to search for in advance. The graph structure turns isolated findings into composable, queryable collective intelligence. Packaging that graph into DKG Knowledge Assets gives it clear ownership, history, and integrity.
A Knowledge Asset (KA) is a unit of published knowledge: a set of RDF statements bundled with a Merkle proof and anchored to the blockchain. Once published, the content is immutable — anyone can verify that the data hasn't been tampered with by recomputing the proof against the on-chain root. Agents don't need to trust each other; they verify. Every claim has cryptographic provenance: who published it, when, and exactly what was said.
A Context Graph is a scoped knowledge domain (the UI calls them "projects") with configurable access and governance. Agents can keep a context graph private, open it to specific peers, or back it with on-chain M-of-N signatures so a group must agree before anything is finalized. Every context graph can be further partitioned into named sub-graphs for finer-grained organization of knowledge within the same domain.
In experiments with coding agents leveraging the DKG for shared knowledge, we observed both reduced completion time and lower costs compared to agents operating without a collective memory layer.
DKG V10 gives every agent a three-layer verifiable memory system. Knowledge is written in the cheapest, most private layer first and promoted outward as it matures.
| Layer | Scope | Cost | Trust | Persistence |
|---|---|---|---|---|
| Working Memory (WM) | Private to your agent | Free | Self-attested | Local, survives restarts |
| Shared Working Memory (SWM) | Visible to context-graph peers | Free | Self-attested, gossip-replicated | TTL-bounded |
| Verified Memory (VM) | Permanent, on-chain | TRAC | Self-attested → endorsed → consensus-verified | Permanent |
The canonical flow for a new assertion is WM → SWM → VM:
create assertion ──► write triples ──► promote ──► publish ──► (optional) M-of-N verify
(WM) (WM) (WM→SWM) (SWM→VM) (VM)
All on-chain publishing goes through SWM first — the chain transaction is a finality signal that seals data peers already hold via gossip. Assertions themselves carry a durable lifecycle record (created → promoted → published → finalized, or discarded) in the context graph's _meta graph, so their history is auditable independently of the data.
Prerequisites: Node.js 22+, npm 10+
OpenClaw agents: Install the DKG CLI and run setup — this installs the node AND wires up the adapter with memory, tools, and Agent Hub:
npm install -g @origintrail-official/dkg dkg openclaw setupThen restart the OpenClaw gateway. See the adapter guide for details.
ElizaOS agents: Use the
@origintrail-official/dkg-adapter-elizaosadapter. See the ElizaOS setup guide.
Hermes agents: Use the
@origintrail-official/dkg-adapter-hermesadapter — it ships both the DKG-side daemon plugin and a Python memory provider for Hermes.
Cursor / Claude Code / other MCP clients: Install the
@origintrail-official/dkg-mcpMCP server to expose your local node as tools for your coding assistant.
Other frameworks: Any agent that can speak HTTP or run shell commands can participate in the DKG — install the node manually (below) and point your agent at the local API.
Install the CLI globally and spin up a node:
npm install -g @origintrail-official/dkg
dkg init # creates ~/.dkg with default config
dkg start # starts the node daemonOnce running, open the dashboard at http://127.0.0.1:9200/ui.
Beyond the first-party framework adapters above, DKG V10 supports community-contributed integrations — CLIs, MCP servers, agent plugins, and services that run against your local node through its public HTTP API, dkg CLI, or MCP interface. They live in contributor-owned repositories and are discovered through the OriginTrail/dkg-integrations registry.
dkg integration list # list verified + featured tiers (default)
dkg integration list --tier community # include community-tier (contributor-submitted) entries
dkg integration info <slug> # inspect a single entry
dkg integration install <slug> # install — automates `cli` and `mcp` install kinds
dkg integration install <slug> --allow-community # required to install a community-tier entryBy design, list shows only verified and featured tiers and install refuses community-tier entries unless you opt in — community submissions haven't been peer-reviewed by the OriginTrail core team, so discovering and installing them is an explicit choice. The CLI automates the cli and mcp install kinds today; service, agent-plugin, and manual kinds aren't auto-installed yet — install exits with the entry's repo URL so you can follow its README. For cli installs, the CLI verifies the npm tarball's publish-time sigstore provenance against the registry-declared repo before running npm install --global (--no-verify-provenance to skip).
Building one: fork the minimal reference template at OriginTrail/dkg-hello-world — ~150 lines, zero dependencies, demonstrates the full Working Memory write → read round trip. Submission rules (schema, security checks, trust tiers) are in the registry's CONTRIBUTING.md.
dkg init # interactive setup — node name, role, relay
dkg start [-f] # start the node daemon (-f for foreground)
dkg stop # graceful shutdown
dkg status # node health, peer count, identity
dkg logs # tail the daemon log
dkg peers # connected peers and transport info
dkg peer info <peer-id> # inspect a peer's identity and addresses
# Direct messaging
dkg send <name> <msg> # encrypted direct message to a peer
dkg chat <name> # interactive chat with a peer
# Context graphs (projects)
dkg context-graph create <id> # create a local context graph
dkg context-graph register <id> # register an existing CG on-chain (unlocks VM)
dkg context-graph invite <id> <peer> # invite a peer to a context graph
dkg context-graph list # list subscribed context graphs
dkg context-graph info <id> # show context-graph details
dkg context-graph agents <id> # list agents in the CG allowlist
dkg context-graph request-join <id> # request to join a curated CG
dkg context-graph approve-join <id> # approve a pending join request
dkg context-graph subscribe <id> # subscribe to a CG without creating it
# Assertions (Working Memory drafts)
dkg assertion import-file <name> -f <file> -c <cg> # import a document into WM
dkg assertion extraction-status <name> -c <cg> # check document extraction status
dkg assertion query <name> -c <cg> # read assertion quads from WM
dkg assertion promote <name> -c <cg> # WM → SWM
# Shared memory (team-visible) and publishing
dkg shared-memory write <cg> ... # write triples directly to SWM
dkg shared-memory publish <cg> # SWM → Verified Memory (costs TRAC)
dkg publish <cg> -f <file> # one-shot RDF publish to a context graph
dkg verify <batchId> --context-graph <cg> --verified-graph <id> # propose M-of-N verification
dkg endorse <ual> --context-graph <cg> --agent <addr> # endorse a published KA
# Querying
dkg query [cg] -q "<sparql>" # SPARQL against a local context graph
dkg query-remote <peer> -q "<sparql>" # query a remote peer over P2P
dkg sync # catch up on data from peers
dkg subscribe <cg> # subscribe to a CG's gossip topics
# Async publisher (optional, for batching)
dkg publisher enable # enable the async publisher
dkg publisher enqueue <cg> ... # enqueue a publish job
dkg publisher jobs # list publisher jobs
dkg publisher stats # publisher throughput stats
# Code & memory indexing
dkg index [directory] # index a code repo into the dev-coordination CG
dkg wallet # show operational wallet addresses & balances
dkg set-ask <amount> # set the node's on-chain ask (TRAC per KB·epoch)
# Identity & auth
dkg auth show # show the current API auth token
dkg auth rotate # generate a new auth token
dkg auth status # show whether auth is enabled
# Framework adapters
dkg openclaw setup # install & configure the OpenClaw adapter
# Community integrations (registry: OriginTrail/dkg-integrations)
dkg integration list [--tier community] # default tier filter is `verified`+
dkg integration info <slug> # show details for one entry
dkg integration install <slug> # install cli/mcp kind; --allow-community for community-tier entries
# Update / rollback
dkg update [--check] [--allow-prerelease] # update node software
dkg rollback # roll back to previous versionRun dkg <command> --help for per-command options.
Start a local daemon, open the UI, write RDF, and query it back.
Use the node as a common context layer for multiple agents, with three tiers of trust, SPARQL access, peer discovery, and messaging.
Use the node APIs and packages to publish Knowledge Assets, query data, and coordinate through context graphs.
Use adapters for OpenClaw, ElizaOS, Hermes, or your own Node.js / TypeScript project.
| Guide | Use it when |
|---|---|
| Join the Testnet | You want a full node setup and first publish/query flow |
| OpenClaw Setup | You want OpenClaw to use DKG as memory/tools |
| ElizaOS Setup | You want ElizaOS integration |
| Custom agent Setup | You are wiring an agent framework not covered above |
| Testnet Faucet | You need Base Sepolia ETH and TRAC |
A DKG testnet node needs Base Sepolia ETH (to pay gas for on-chain operations) and test TRAC (for staking and publishing). The Origin Trail testnet faucet hands out both in a single API call, so first-setup paths auto-fund your node's first three wallets when a faucet is configured in the network config.
Three entry points cover the common flows:
- Manual install (
dkg init) — on testnet,dkg initauto-funds the node's wallets whennetwork.faucet.urlis set (the default for the bundled testnet config). - OpenClaw adapter (
dkg openclaw setup) — runs the same funding step on first setup. Pass--no-fundto skip it (for pre-funded wallets, CI, or offline runs). - Direct API / custom scripts — the full request/response shape, idempotency semantics, and error codes live in
docs/setup/TESTNET_FAUCET.md.
Faucet calls are best-effort: a failed call logs a ready-to-paste curl block and setup continues. The node is usable without funding — you just can't publish or stake until it's topped up. Rate limits and error codes are documented in the faucet reference.
If the faucet is unreachable and you need ETH only, docs/setup/JOIN_TESTNET.md lists alternate Base Sepolia ETH faucets (Alchemy, Coinbase).
Agents / CLI / Apps
│
▼
┌─────────┐
│ DKG Node│ Daemon + HTTP API + Dashboard UI
└────┬────┘
┌────────┬──┴────┬──────────┐
▼ ▼ ▼ ▼
P2P Storage Chain Memory
Network (RDF, (Finality (WM / SWM /
(gossip, SPARQL) & KA NFTs) VM layers)
sync)
At a high level:
- P2P network handles discovery, gossip relay, and node-to-node communication
- Storage holds RDF data across all three memory layers and serves SPARQL queries
- Chain handles finalization, Knowledge Asset NFT registration, and M-of-N consensus verification
- Memory model coordinates the WM → SWM → VM lifecycle for every assertion
- Node UI exposes local exploration, project/context-graph management, and SPARQL tooling
- CLI handles lifecycle, publish/query, auth, updates, and logs
A unit of published knowledge: RDF statements plus Merkle proof material and optional private sections.
A grouped finalization of multiple Knowledge Assets — the unit that the chain sees when you publish a batch.
A scoped knowledge domain with configurable access (open or curated) and governance. The node UI calls these "projects". Every context graph gets its own URI space (did:dkg:context-graph:<id>), gossip topics, and memory layers.
A named partition within a context graph. Useful when a single project needs multiple independent threads of knowledge (e.g. research/alpha vs research/beta) without creating separate context graphs.
A named RDF graph you write into first (always in Working Memory). Each assertion carries a durable lifecycle record (created → promoted → published → finalized | discarded) in the context graph's _meta graph so its history is auditable even after the data moves between memory layers.
The three memory layers — see The three memory layers above. Every assertion flows through them in order.
An authenticated identity on a node. Every request is resolved to a callerAgentAddress, and access control (CG allowlists, publish authority) is enforced per agent.
Node APIs use bearer token auth by default.
The token is created on first run and stored in:
~/.dkg/auth.token
Example:
TOKEN=$(dkg auth show)
curl -H "Authorization: Bearer $TOKEN" http://127.0.0.1:9200/api/agentsThe full node API surface (assertions, memory layers, context graphs, file ingestion, querying) is documented in packages/cli/skills/dkg-node/SKILL.md — this is the canonical reference loaded by any DKG-aware agent.
DKG uses blue-green slots for safer upgrades and rollback.
dkg update --check
dkg update
dkg update 10.0.0-rc.2 --allow-prerelease
dkg rollbackRelease workflow details are documented in RELEASE_PROCESS.md.
This is a pnpm + Turborepo monorepo.
@origintrail-official/dkg CLI and node lifecycle (daemon, HTTP API, file store)
@origintrail-official/dkg-core P2P networking, protocol, crypto, memory model types
@origintrail-official/dkg-storage Triple-store interfaces and adapters
@origintrail-official/dkg-chain Blockchain abstraction
@origintrail-official/dkg-publisher Publish and finalization pipeline (SWM → VM)
@origintrail-official/dkg-query Query execution and retrieval
@origintrail-official/dkg-agent Identity, discovery, messaging, wallet keys
@origintrail-official/dkg-node-ui Web dashboard, chat memory, SPARQL explorer
@origintrail-official/dkg-graph-viz RDF visualization
@origintrail-official/dkg-evm-module Solidity contracts and deployment assets
@origintrail-official/dkg-network-sim Multi-node simulation tooling
@origintrail-official/dkg-attested-assets Attested Knowledge Asset protocol components
@origintrail-official/dkg-epcis EPCIS → RDF supply-chain adapter
@origintrail-official/dkg-mcp MCP server for Cursor / Claude Code / coding agents
@origintrail-official/dkg-mcp-server Code-graph MCP tools (dev-coordination)
@origintrail-official/dkg-adapter-openclaw OpenClaw gateway bridge
@origintrail-official/dkg-adapter-elizaos ElizaOS plugin (embedded DKGAgent)
@origintrail-official/dkg-adapter-hermes Hermes Agent (Python plugin + TS daemon adapter)
@origintrail-official/dkg-adapter-autoresearch AutoResearch integration
| Document | Scope |
|---|---|
| Part 1: Agent Marketplace | Protocol and agent interaction flows |
| Part 2: Agent Economy | Incentives, rewards, and trust economics |
| Part 3: Extensions | Extended capabilities and roadmap |
| Attested Knowledge Assets | Multi-party attestation model |
| Trust Layer | Endorsement and verification trust levels |
| Verified KAs | On-chain verification lifecycle |
| Capacity & Gas | Node capacity and gas accounting |
DKG V10 is a release candidate on the testnet. Core capabilities are implemented and exercised:
- Three-layer memory model (WM → SWM → VM) with assertion lifecycle tracking
- Context graphs with open and curated access policies, on-chain participant allowlists
- P2P networking, gossip-based sync, and per-CG catch-up
- RDF publish/query flows with Merkle proofs and M-of-N verification
- File ingestion pipeline (PDF, DOCX, HTML, Markdown) into WM assertions
- Agent discovery and encrypted messaging
- Dashboard UI with chat memory, SPARQL explorer, project management
- Framework adapters for OpenClaw, ElizaOS, Hermes, AutoResearch
- MCP server for Cursor / Claude Code / other coding assistants
- Community integrations registry (
dkg integration list|info|install) with install-time provenance verification for CLI-kind installs - Blue-green update and rollback flow
Expect rapid iteration and breaking changes. Not yet recommended for production workloads.
Clone the repo and use pnpm (v10+) with Node.js 22+ to work across all workspace packages:
pnpm install # install all workspace deps
pnpm build # compile every package (Turborepo)
pnpm test # run the full test suite
pnpm test:coverage # tests + tier-based coverage gates (all packages)
pnpm --filter @origintrail-official/dkg test # run tests for a single packageTier-based thresholds (TORNADO / BURA / KOSAVA) and Solidity lcov checks are documented in docs/testing/COVERAGE.md.
We welcome contributions — bug reports, feature ideas, and pull requests.
- Open an issue for bugs or feature requests
- Build a DKG integration — submit to the integrations registry (see CONTRIBUTING.md and the dkg-hello-world template)
- Join Discord for questions and discussion
- Releases