Agent88 integration docs · OriginTrail DKG V10 bounty

Build the open memory layer for AI.

This is an Agent88 integration blueprint for the OriginTrail DKG V10 bounty program: connect Hermes, OpenClaw, and Agent88 research/CRM workflows to DKG Working Memory and Shared Memory so AI agents can write, recall, share, and eventually verify knowledge without locking it inside one vendor platform.

Integration blueprint Build plan Bounty fit

Round 1

Working Memory + Shared Memory integrations

Up to 10,000 TRAC

per accepted contribution, tiered by impact

50,000 TRAC

Round 1 pool cap, within the broader 150,000 TRAC program

Priority fit

OpenClaw, Hermes, autoresearch agents, LLM-Wiki style workflows

Core proposal

Agent88 DKG Memory Bridge

A lightweight integration that turns Agent88 agent work into DKG-native memory. Every meaningful output — research note, proposal draft, meeting summary, CRM insight, code-review finding, source citation, workflow decision — is written first to private Working Memory, then promoted to team Shared Memory when it is useful, safe, and reusable.

Memory layers

Use the DKG trust gradient exactly as intended.

Working Memory

Private agent/node draft layer

Store raw notes, turn summaries, extracted research, CRM observations, and source-backed claims before sharing.

Shared Memory

Team/context-graph layer

Promote useful artifacts for other Agent88 agents and operators to query, reuse, contest, and build on.

Verified Memory

On-chain anchored layer

Not the main Round 1 target, but every data shape should preserve provenance so later publication to Verified Memory is natural.

Architecture

From agent artifact to shared AI memory.

The integration should be boring and reliable: capture useful agent output, normalize it into entities and RDF quads, write it to Working Memory, query it back, then promote only when the artifact deserves to become shared team substrate.

Capture

Hermes / OpenClaw / Telegram / research agents produce turns, artifacts, decisions, tasks, reports, and source links.

Normalize

Agent88 Memory Bridge maps each artifact into DKG-friendly entities: Agent, Source, Claim, Artifact, Decision, Task, Workflow, Client, and Context Graph.

Write to Working Memory

Use DKG MCP, CLI, or local HTTP API to create assertions and write RDF quads with provenance, timestamps, source URLs, author agent, and privacy status.

Recall

Agents query Working Memory with dkg_memory_search or SPARQL before repeating research, drafting follow-ups, or creating workflow proposals.

Promote to Shared Memory

Human-approved or high-confidence artifacts are promoted from private draft memory into team-visible Shared Memory.

Prepare for Verified Memory

Stable, source-backed, non-sensitive outputs can later be published or requested for publication as Knowledge Assets / Knowledge Collections.

Agent88 surfaces

Where the bridge should attach first.

Hermes Agent

Use dkg hermes setup or DKG MCP tools so completed turns, distilled memories, research notes, and workflow outputs can be written into DKG memory.

OpenClaw

Route build orchestration artifacts, Kanban decisions, worker outputs, code-review findings, and deployment notes into project Context Graphs.

Agent88 NIA / CRM

Capture relationship dossiers, intro decisions, prospect workflow pains, and follow-up actions as private Working Memory, then promote qualified opportunities to Shared Memory.

Research ingestion

Turn PDFs, links, RSS, arXiv, GitHub issues, PR reviews, and market notes into source-backed entities that agents can query across projects.

Website docs

Publish sanitized integration briefs, design docs, and public examples while keeping client-sensitive memory private.

Data model

Keep the ontology small enough to ship.

Round 1 should not drown in schema design. Use a minimal vocabulary that preserves provenance, source references, status, privacy, and promotion readiness.

agent88:Artifact

A generated report, proposal, transcript summary, code review, research digest, or BD note.

agent88:Source

Original URL, PDF, spreadsheet, Telegram message, meeting transcript, GitHub issue, or human note.

agent88:Claim

Atomic statement extracted from a source, with confidence, evidence, and authoring agent.

agent88:Workflow

A repeatable business process such as proposal generation, reporting, funding navigation, CRM follow-up, or research ingestion.

agent88:Decision

A human or agent-assisted decision with rationale and approval status.

agent88:PromotionCandidate

A Working Memory artifact that is ready for Shared Memory, and later possibly Verified Memory.

Implementation plan

Six steps to a bounty-ready integration.

Set up DKG V10 locally

Install the CLI, start the node, configure Hermes/MCP, and verify the node UI at localhost:9200.

Create Agent88 context graphs

Start with agent88-research, agent88-crm, agent88-code, and agent88-public-docs. Use subgraphs such as chat, sources, claims, decisions, tasks, crm, code, and meta.

Implement write path

For each Agent88 artifact, create a Working Memory assertion, write RDF quads, and query it back as a round-trip test.

Implement recall path

Expose memory_search and SPARQL query helpers so agents can retrieve prior claims, tasks, decisions, and sources before drafting new work.

Implement promotion path

Add human approval gates before promoting from Working Memory to Shared Memory. Do not auto-publish to Verified Memory in Round 1.

Package as bounty submission

Publish a contributor-owned repo, package with provenance, tests against a local DKG node, design brief, demo video, security notes, and registry PR.

Operator quickstart

DKG local setup and first memory round-trip.

npm install -g @origintrail-official/dkg
dkg hermes setup --no-fund
dkg mcp setup --yes --no-fund
dkg status
dkg context-graph create agent88-research
dkg assertion import-file event-bd-workflow -f ./docs/event-bd-workflow.md -c agent88-research
dkg assertion promote event-bd-workflow -c agent88-research
dkg query agent88-research -q "SELECT ?s ?p ?o WHERE { ?s ?p ?o } LIMIT 20"

Note: use --no-fund for dry local setup, or fund testnet wallets when testing publish flows. Round 1 should prove Working and Shared Memory first; Verified Memory should be treated as a later promotion path, not the default write path.

In scope

• Writes to or reads from DKG V10 Working Memory or Shared Memory through public interfaces: node HTTP API, dkg CLI, or MCP.
• Connects memory to real agent workflows: Hermes, OpenClaw, autoresearch, research ingestion, GitHub/PR memory, CRM/NIA, or LLM-Wiki-style knowledge work.
• Uses v10 vocabulary: Context Graph, Integration, Curator, Entity, Knowledge Asset, Knowledge Collection, SHARE, PUBLISH, Projects.
• Documents how artifacts can mature toward Verified Memory and context-oracle consumption later.

Avoid

• Verified-Memory-only or chain-anchoring-only work for Round 1.
• UI voting, endorsement buttons, or consensus screens; v10 expects conversational consensus through agents.
• DKG v9-only work or Situation Room-style v9 apps without v10 Working/Shared Memory.
• Importing internal monorepo packages or patching the DKG node source. Build against public API, CLI, or MCP surfaces.
• Bypassing Curator authority for SHARE or PUBLISH operations.

Submission package

What Agent88 should prepare for the bounty.

• Contributor-owned GitHub repo for the integration, not a patch to the DKG monorepo.
• Published npm package or equivalent verifiable registry package with provenance.
• Design brief covering problem, target user, memory layers touched, v10 primitives used, LLM-Wiki / autoresearch fit, terminology, and promotion path.
• Working demo video showing write → recall → promote from Working Memory to Shared Memory.
• Integration tests against a local DKG V10 node.
• Security notes: credentials, data egress, write authority, Curator authority, privacy boundaries, and no dynamic remote code loading.
• Registry PR against OriginTrail/dkg-integrations, pinned to package version and commit.

Recommended Agent88 angle

Ship a Hermes/OpenClaw research-memory bridge first.

This is the highest-fit Agent88 route: it uses our real operating cockpit, captures real long-horizon agent work, proves Working and Shared Memory, and creates a credible path toward Verified Memory and context oracles without overbuilding chain UX in Round 1.