ckg-agentforce 0.2.0

Salesforce AgentForce as a traversable knowledge graph — 40 nodes, Einstein Trust Layer, Resolution Criteria, $2/resolution billing path. 11× fewer tokens than RAG. MCP-native.

ckg-agentforce

Salesforce AgentForce as a traversable knowledge graph.
40 nodes · 5 tools · 2 resources · 1 resource template · 2 prompts · MCP-native

AgentForce charges $2 per autonomous resolution. Every resolution requires the agent to correctly traverse: Einstein Agent → Resolution Criteria → Audit Trail → Einstein Trust Layer → Policy Enforcement. Wrong inference means retry, CSAT impact, and a $2 charge with no result.

Vector RAG retrieves ~2,982 tokens of Salesforce docs. The agent infers what matters. Confident. Possibly wrong.

This graph declares what the agent should already know.

pip install ckg-agentforce
uvx ckg-agentforce          # run as MCP server

---

Knowledge Graph — 40 nodes

                        AgentForce Platform
                        ╱       │        ╲
              Einstein Agent  Data Cloud  Model Selection
             ╱    │    │    ╲      │            │
    Agent Topic   │  Grounding  Salesforce   Reasoning Engine
         │        │     │       CRM        ╱    │    ╲
   Agent Action  Resolution  Knowledge   NVIDIA  Token  Context
   ╱  ╱  ╱  ╲   Criteria    Base        NIM    Budget  Window
  Flow Apex Mul. Ext.   ╲
  Act. Act. Act. Act.   Autonomous Resolution ←── Audit Trail
                            ($2 / event)               │
                                │              Einstein Trust Layer
                           Agent Metrics       ╱   │     │    ╲
                           Handoff to     Data   Zero  Policy  Compliance
                             Human       Masking  DR   Enforce  Rules

Node taxonomy (40 total)

Type	Count	Key concepts
CAPABILITY	7	Flow · Apex · External · MuleSoft · Standard · Custom Actions · Grounding
COMPLIANCE	4	Audit Trail · Zero Data Retention · Policy Enforcement · Compliance Rules
AGENT	4	Einstein Agent · Service Agent · Sales Agent · Marketing Agent
CONCEPT	3	Agent Topic · Resolution Criteria · Conversation State
WORKFLOW	3	Handoff to Human · Omni-Channel Routing · Multi-Agent Orchestration
CONFIG	3	Agent Instruction · Token Budget · Model Selection
SECURITY	2	Einstein Trust Layer · Data Masking
MECHANISM	2	Grounding · Retrieval Augmented Generation
PLATFORM	2	AgentForce Platform · Data Cloud
ARTIFACT	2	Knowledge Base · Prompt Template
Other	8	BILLING · ALGORITHM · DATASOURCE · ENGINE · RUNTIME · MONITORING · CONSTRAINT · TOOL

---

Install

pip install ckg-agentforce

Requires Python ≥ 3.10 · mcp >= 1.0.0

---

Setup

Claude Desktop

{
  "mcpServers": {
    "agentforce": {
      "command": "uvx",
      "args": ["ckg-agentforce"]
    }
  }
}

Claude Code CLI

claude mcp add agentforce uvx ckg-agentforce

Cursor / Windsurf

{
  "mcpServers": {
    "agentforce": {
      "command": "uvx",
      "args": ["ckg-agentforce"]
    }
  }
}

Verify

list_concepts()
resolution_path()
query_ckg('Autonomous Resolution', 4)

---

Tools

list_concepts()

All 40 AgentForce concepts, grouped by type.

ckg-agentforce — 40 concepts:
  AgentForce Platform   [PLATFORM]
  Einstein Agent        [AGENT]
  Einstein Trust Layer  [SECURITY]
  Autonomous Resolution [BILLING]
  ...

search_concepts(query)

Find concepts by keyword.

search_concepts("resolution")   # → Resolution Criteria · Autonomous Resolution
search_concepts("trust")        # → Einstein Trust Layer · Zero Data Retention
search_concepts("action")       # → all 7 action types

query_ckg(concept, depth=3)

Traverse any concept — prerequisites upstream, dependents downstream. Every relationship traces to a declared Salesforce doc.

query_ckg('Autonomous Resolution', 4)

## Autonomous Resolution  ·  AgentForce CKG
Type: BILLING · Depth: 4 hops

Prerequisites
  - [REQUIRES:1.00] Resolution Criteria     (CONCEPT)
  - [REQUIRES:1.00] Audit Trail             (COMPLIANCE)
    - [REQUIRES:1.00] Einstein Trust Layer  (SECURITY)
      - [REQUIRES:1.00] AgentForce Platform (PLATFORM)
      - [REQUIRES:1.00] Reasoning Engine    (ENGINE)

Builds toward
  - [REQUIRES:0.95] Handoff to Human  (WORKFLOW)
  - [REQUIRES:0.90] Agent Metrics     (MONITORING)

This traversal: ~269 tokens · RAG equivalent: ~2,982 tokens · 11× compression

get_prerequisites(concept)

Full ordered upstream chain.

get_prerequisites('Custom Actions')
# → AgentForce Platform → Agent Topic → Agent Action → Apex Action → Custom Actions

resolution_path()

The exact $2/resolution billing traversal — every hop declared.

Einstein Agent [AGENT]
  requires: AgentForce Platform · Reasoning Engine · Einstein Trust Layer

Resolution Criteria [CONCEPT]
  requires: Einstein Agent

Autonomous Resolution [BILLING]
  requires: Resolution Criteria · Audit Trail

Audit Trail [COMPLIANCE]
  requires: Einstein Trust Layer

Policy Enforcement [COMPLIANCE]
  requires: Einstein Trust Layer

Einstein Trust Layer [SECURITY]
  requires: AgentForce Platform

---
BEFORE (RAG): 2,982 tokens · model infers · confident, possibly wrong
AFTER  (CKG):  269 tokens · 4 declared hops · auditable by design

---

Resources

Two static resources readable by any MCP client.

URI	Contents
agentforce://nodes	All 40 concepts grouped by taxonomy type
agentforce://resolution-chain	The $2/resolution billing path with declared edges

---

Resource Templates

Parameterized access to any concept in the graph.

Template	Returns
agentforce://concept/{concept}	Prerequisites + dependents for any named concept

Example: read agentforce://concept/Einstein Trust Layer

# Einstein Trust Layer  [SECURITY]

## Prerequisites (1)
  - AgentForce Platform

## Dependents (6)
  - Data Masking · Audit Trail · Zero Data Retention
  - Grounding · Einstein Agent · Policy Enforcement

---

Prompts

show_burn

Walks the resolution path and renders a before/after: RAG 2,982 tokens vs CKG 269 tokens, with the exact hop path.

Steps it runs: resolution_path() → query_ckg('Autonomous Resolution', 4) → query_ckg('Einstein Trust Layer', 3)

map_agentforce_stack

Maps the full platform from Einstein Agent to billing, then renders an interactive D3.js force-directed graph.

Node colors: AGENT=teal · SECURITY=red · BILLING=amber · CAPABILITY=blue · PLATFORM=slate

---

Benchmark

Metric	RAG	GraphRAG	CKG
F1	0.123	0.120	0.471
Tokens / query	2,982	—	269
Traceability	model infers	partial	declared edge

v0.6.2 · 7,928 queries · 45 domains
Dataset: huggingface.co/datasets/danyarm/ckg-benchmark
Paper: github.com/Yarmoluk/ckg-benchmark/blob/main/paper/main.pdf

---

Enterprise stack

97 domains across NVIDIA, finance, regulatory, and enterprise AI. Remote MCP connector — no install, paste one URL into claude.ai or Cursor.

graphifymd.com/pro/ — $99/mo

---

EVAL

benchmark: ckg-benchmark v0.6.2
dataset: huggingface.co/datasets/danyarm/ckg-benchmark
benchmarked: false
rag_baseline_f1: 0.123
graphrag_baseline_f1: 0.120
mean_tokens: 269
paper: github.com/Yarmoluk/ckg-benchmark/blob/main/paper/main.pdf

---

Graphify.md · patent pending · graphifymd.com