ClearPoint Logic

Three products. One trust fabric. One partner network.

ClearPoint Logic is built in layers. Anchor at the bottom — the trust fabric that signs every artifact. Agent Core in the middle — where agents actually run, across the major frameworks. CPL Agents on top — the signed agent packages that ship into your environment.

Talk to engineering

How the platform is layered.

The same architecture diagram our engineers walk customers through on day one.

01SurfacesProducts customers interact with
  • StudioBuild
    Compose, test, and ship agents.
  • NexusDistribute
    Catalog and partner network for the AI workforce.
  • MeridianGovern
    System of record for the AI workforce.
02Trust fabricIdentity, policy, and signed evidence
Anchor
Trust fabric
Cert. Standard
Open spec
AI BOM
Per-artifact
Agent Passport
Identity
Signed Evidence
Per-run
03Agent runtimeWhere agents run, across frameworks
Agent Core
Agent runtime
Framework
ADK Go
Framework
LangChain
Framework
Anthropic SDK
Framework
Microsoft AF
04Agent codeSigned packages that load into Agent Core
CPL Agents
Signed agent packages
  • invoice-triage@1.4
  • lead-enrichment@2.1
  • contract-redline@0.8
  • ticket-router@3.0
  • doc-summarizer@1.2
The CPL Agents tier shows example signed packages.
Layer 02 · Anchor

The trust fabric.

Anchor is the layer everything in the platform is signed against. It ties people, agents, models, prompts, tools, datasets, and runs together into a single signed record — auditable end-to-end, exportable to your evidence system.

Certification Standard
An open spec for what "governed" means at the artifact level. Every claim points to it.
AI Bill of Materials
A signed manifest per agent: every model, prompt, tool, dataset, dependency, and version.
Agent Passport
A vendor-neutral identity record. The same passport works on Microsoft, Google, custom, BYOA.
Signed Evidence
Every run produces a signed trace — inputs, decisions, approvals, outputs — retained per policy.
AgentPassportAI BOMModelToolRunOperatorDataset
CoreIdentitySpecificationEvidence
Frameworks (pluggable)
ADK Go
LangChain
Anthropic SDK
MS Agent FW
Agent Core runtime
Lifecycle
deploy · canary · rollback
Approvals
role · dual · reversible
Audit
signed traces · streaming
Memory layer
Cloud SQL · pgvector · tenant-partitioned
Kill switches
agent · tool · tenant
Anchored to Anchoridentity · evidence
Layer 03 · Agent Core

Where agents run.

Agent Core runs your agents — whichever framework they were written in. ADK Go, LangChain, Anthropic Agent SDK, Microsoft Agent Framework — all of them share the same identity, lifecycle, approvals, evidence, and memory.

  • Multi-framework runtime
    Run agents written for ADK Go, LangChain, Anthropic SDK, or Microsoft AF — same lifecycle, same telemetry.
  • Lifecycle management
    Versioning, blue/green, canary, rollback. Promotion is a policy decision, not a deploy.
  • Approval routing
    Human-in-the-loop checkpoints with role-based routing, dual control, and reversible actions.
  • Audit pipeline
    Every tool call, prompt, and decision streamed as signed traces — retained per policy.
  • Memory layer
    Cloud SQL with pgvector, tenant-partitioned. Per-agent retention and PII redaction at ingest.
  • Kill switches
    Per-agent, per-tool, per-tenant pause and revoke — enforced at the runtime, not by configuration.
Layer 04 · CPL Agents

The agent code layer.

CPL Agents are the actual agent implementations — code, prompts, tool bindings, and AI BOM. They’re published as signed packages from Studio, certified against the Anchor standard, and load at runtime into Agent Core or a customer-deployed Meridian instance.

Multi-framework: an agent can be authored in ADK Go, LangChain, Anthropic SDK, or Microsoft Agent Framework. Same publishing pipeline, same passport, same evidence. The framework is an implementation detail; the contract with the platform is not.

  • signed packages
  • AI BOM included
  • framework-agnostic
  • reproducible builds
  • reversible deploy
invoice-triage@1.4.0 · package.json
{
  "name": "invoice-triage",
  "version": "1.4.0",
  "framework": "anthropic-sdk",
  "passport": "cpl:passport:8f2a91c4",
  "bom": {
    "models":   ["claude-3-5-sonnet@20251022"],
    "tools":    ["sap.lookup", "erp.write", "pii.redact"],
    "datasets": ["finops.invoices.v3"],
    "prompts":  ["sha256:a4f2…"]
  },
  "scopes":     ["read:invoices", "write:erp"],
  "policy":     "envelope://acme/finance/tier-1",
  "signature":  "anchor:sig:rsa256:c7d1…"
}
Build
Studio · reproducible
Deploy
Agent Core · Meridian
Cross-vendor by design

Govern every agent. Not just one ecosystem.

Vendor governance products only see their own ecosystem. Meridian sits above them — one passport, one policy envelope, one evidence layer across every place an agent runs.

  • Microsoft
  • Google
  • Salesforce
  • ServiceNow
  • AWS
  • Anthropic
  • OpenAI
One passport · one policy envelope · one evidence layer

Google First. Anthropic Partner.

  • Google First
    Cloud · runtime · model

    The platform is GCP-native. Agent Core runs on the Google Cloud control plane; ADK Go is a first-class runtime path; Gemini is integrated as a reasoning option alongside Anthropic Claude.

    • GCP-native control plane
    • ADK Go runtime in Agent Core
    • Gemini integrated reasoning
    • Cloud SQL + pgvector memory
  • Anthropic Partner
    Reasoning · MAP · network

    Claude is the primary reasoning model for CPL Agents. We integrate with the Managed Agent Platform and align with the Anthropic Partner Network for distribution and reference patterns.

    • Claude as primary reasoning
    • Managed Agent Platform integration
    • Anthropic Partner Network alignment
    • Anthropic Agent SDK runtime support

Want to go deeper?

Read the strategy brief for the architectural philosophy, or talk to engineering for a technical walkthrough of the platform layers.

Talk to engineeringRead the strategy brief