Skip to content
all frameworks
09 · Framework

Agent Stack Map

Harness × model × tools → eight configurations → a portfolio
Three axes, chosen independently
01Harnessloop, context, sandbox
02Modelquality, place of inference
03Toolsactions, identity, execution
Eight configurations on the data-path and authority axes
Corethe working center
2Own harness + external API4Vendor agent + internal gateway
Scaleonly after evals
6Multi-model router
Restricteddata stays inside
1Self-hosted, air-gapped3Ready client + local model
High riskplanning split from execution
7External planner + internal executor
Sandboxno work data or authority
5Full SaaS8Personal stack + external MCP
No universal ranking: status is assigned per scenario and data class.

The argument 'our own client on our own model versus a vendor agent over an API' conflates independent decisions. Harness, model and tools are chosen separately: each axis carries its own data path, cost, failure point and vendor lock-in.

Eight repeatable configurations sit on the data-path and authority axes — from a self-hosted stack in an air-gapped perimeter to a personal setup wired to external MCP servers. They carry no universal ranking: air-gap demands the first, pilot speed leads to the fifth, corporate actions to the fourth, high risk to the seventh.

So the organizational outcome is not a winner but a portfolio with statuses: a working core, the next step after evals, restricted perimeters, high-risk execution and a sandbox. The company owns the gateways, identity, policy and traces, while clients and models stay replaceable.

How to use this model
01

Decompose your stack along three axes: harness, model, and tools — each is chosen independently.

02

Answer three questions: does data leave the perimeter, are internal actions required, do you have scale and evals.

03

Assign every configuration a portfolio status per scenario and data class, not once for the whole company.

Source talks
Share