PARTNER AgentFoundry  ×  Cisco AI-Defense

The IDE Cisco-AI-Defense customers ship agents from.

Multi-model. Governed by TrustModel AGP. Observable through Galileo. One dev surface that respects the policy enforcement Cisco already sells into the enterprise.

Email Karl Mehta Read the architecture
Why Cisco · Why now

Three reasons this partnership matters

Cisco's AI-Defense bet is about giving enterprises a way to deploy agents without losing control of them. AgentFoundry is the developer surface that bet is missing.

1

Galileo aligns Cisco with TrustModel-style observability

Cisco's Galileo acquisition standardizes AI-Defense customers on the same trace format, span model, and judge-based eval discipline that TrustModel was built around. The connector is already in production — galileo_connector Django app in aurora-gateway.

2

AI-Defense customers need a dev surface that respects governance

Today, the developer's IDE is the weakest link in the AI-Defense story: agents are built in tools that have zero awareness of the policies enforced at the network edge. AgentFoundry makes governance a first-class IDE primitive — the agent is born with policy attached.

3

AgentFoundry ships with TrustModel AGP built in

Every agent built or run in AgentFoundry flows through a TrustModel AGP edge sidecar. Score, governance attestations, and telemetry are emitted in the same OTLP shape Galileo and Cisco AI-Defense already speak — no new integration tax for the customer.

Architecture

Three layers. One governance stack. All speaking OpenTelemetry.

This mirrors the 3-layer model from the TrustModel + Cisco AI-Defense integration architecture we delivered earlier this quarter. AgentFoundry plugs into each layer as the developer-facing surface.

Layer 1
Protect
Cisco AI-Defense at the network and policy edge. Outbound calls, prompt injection, exfiltration, model abuse — enforced where Cisco already lives. AgentFoundry's agent runs inherit those policies because every model call routes through the customer's gateway, not the developer's laptop. AgentFoundry surface: AGP edge sidecar registers the agent's identity, intended model set, and policy class with Cisco's gateway before first run.
Layer 2
Observe
Galileo for trace, span, and per-step evaluation. Hallucination, toxicity, PII, context relevance, factual consistency at the span level. AgentFoundry emits OTLP traces in the format Galileo expects, so every run from the IDE — local or deployed — is already shaped for Galileo ingest. AgentFoundry surface: galileo.init() wired by default in AgentFoundry's agent scaffolds (Claude Agent SDK, Cursor SDK, Microsoft Agent Framework, Google ADK).
Layer 3
Assure
TrustModel AGP for the 0-100 TrustScore, attestations, and compliance backplane. Galileo traces and AI-Defense policy events normalize through GalileoTransformerTelemetryEvalInput, scored across 10 dimensions, signed as an attestation, surfaced as an audit-ready record. AgentFoundry surface: TrustScore renders inline next to every agent in the IDE; failing dimensions block deploy until reviewed.
Shipped · not vapor

What is already in production

None of this is "if we did a partnership." These pieces exist today and are live with customers.

Shipped

Galileo connector in TrustModel (aurora-gateway)

galileo_connector Django app — REST pull from Galileo project trace DB, GalileoTransformer normalization, MCP forwarding, OTLP push. Live in PR #192.

Shipped

AgentFoundry agent-fleet IDE with built-in AGP edge sidecar

Multi-project, multi-model IDE where every agent runs behind a local AGP sidecar. Policy, trace, and TrustScore are emitted on every run with no app-side instrumentation.

Shipped

TrustModel AGP backplane — TRUS-840 + TRUS-1000

SKU 3 Agent Governance is the platform layer the sidecar reports into. TRUS-1000 (Helm chart) closes the in-VPC deploy gap Fortune 500 enterprises need on day one.

Shipped

TelemetryTraceStrategy + vendor-agnostic transformers

TelemetryEvalInput normalizes Galileo, W&B Weave, Arize, LangSmith, and OTel traces into one shape. Cisco AI-Defense events plug in via the same path.

The motion

A joint enterprise bundle for agent-development customers

Three products, one story. Each owns its layer. Each amplifies the others.

Cisco AI-Defense

Network & policy enforcement at the edge. The customer's existing Cisco footprint becomes the runtime perimeter for every agent built in AgentFoundry.

TrustModel AGP

Agent-governance backplane. TrustScore, attestations, compliance mapping (NIST AI RMF, EU AI Act, ISO 42001). Galileo telemetry feeds it directly.

AgentFoundry

The developer IDE that bakes both in. The agent is born governed, born observable, born inside the Cisco perimeter — no retrofit at deploy time.

Co-sell motion: Cisco AI-Defense lands the enterprise, TrustModel AGP secures the agent layer, AgentFoundry becomes the standard IDE the customer's developers build on. One contract, three vendors, zero integration work for the buyer.

About

About AgentFoundry

AgentFoundry is the IDE for the agent era. 30-second setup, multi-project tabs, multi-model AI (Claude, Codex, Gemini, Aider, Copilot) side-by-side, first-class scaffolds for Claude Agent SDK, Cursor SDK, Microsoft Agent Framework (Azure), and Google ADK (Vertex). Agent-fleet management is built in — every agent the developer ships runs behind a local TrustModel AGP edge sidecar that handles policy, trace, and TrustScore on every invocation.

Token-economics-aware multi-model routing means the right task goes to the right model at the right cost. The IDE replays any production agent run, forks it back to current code, and re-runs it under governance. AgentFoundry is built by Predixtions Inc., the team that built TrustModel — so the IDE and the governance backplane share a roadmap, not just an API.

Let's talk.

If Cisco AI-Defense is the perimeter, TrustModel AGP is the agent backplane, and AgentFoundry is the IDE — there's a real co-sell motion here. Happy to walk through architecture, customer overlap, or a joint pilot.

Email Karl Mehta
Karl Mehta, Founder/CEO · knm@predixtions.com · github.com/karlmehta