Question 1

What does Forgebench govern for developers using AI?

Accepted Answer

What does Forgebench govern for developers using AI?

The API keys your organization issues. Every call made with a Forgebench credential is routed through one path, checked against a ceiling and a rate limit, scanned for what leaves your perimeter, and recorded — with the cost attributed to the developer, team or feature the credential belongs to.

Can you cap what a developer spends on Claude Code or GitHub Copilot?

Which coding assistants do you support?

Do developers have to install anything?

What leaves a developer's machine?

Can developers be measured individually?

How do you connect spend to output?

Do I have to give every developer their own credential?

What happens when a developer hits a ceiling?

Does this replace my model gateway or observability stack?

Do guardrails block or delay a response?

Is it self-hosted?

Can we govern agents in production too?

Can Forgebench govern AI usage across multiple model providers?

Can Forgebench work with AI infrastructure we have already built?

Can Forgebench work with AWS Bedrock?

Can Forgebench work with Azure OpenAI Foundry?

How does Forgebench integrate with our existing SSO and RBAC?

What happens if an application calls an AI provider directly instead of going through Forgebench?

Can we make Forgebench the mandatory path for AI calls?

Can we restrict which models a developer or team can use?

Question 2

Can you cap what a developer spends on Claude Code or GitHub Copilot?

Accepted Answer

No. Those run on your company's own vendor accounts and seat subscriptions, so the spend never passes through Forgebench and there's nothing for us to hold. We integrate read-only and report what each person and team is spending, alongside the work it produced. Caps, rate limits and guardrails apply to the credentials you issue through Forgebench.

Question 3

Which coding assistants do you support?

Accepted Answer

Claude Code, Codex, GitHub Copilot and AWS Kiro — which between them cover most of what enterprises have deployed. The integrations are read-only and use the accounts your company has already issued.

Question 4

Do developers have to install anything?

Accepted Answer

For Claude Code, no — it's a managed-settings file pushed through your MDM. Codex and the working-style profile sync through a small CLI that runs at session end. Jira and GitHub are read-only connections made once, centrally.

Question 5

What leaves a developer's machine?

Accepted Answer

Counts, scores and model names. Prompt text and file contents never do.

Question 6

Can developers be measured individually?

Accepted Answer

That's your choice, and it's configurable rather than assumed. There's an aggregate-only mode with a minimum group size, team-lead scoping so a lead sees only their own team, and a personal view where each developer sees their own record.

Question 7

How do you connect spend to output?

Accepted Answer

A session ties to its branch, the branch to its pull request, and the pull request to the Jira issue it closed. That gives you cost per commit, per pull request and per 1K lines, and tokens spent against story points closed. Sessions without Git context are excluded from those figures rather than estimated.

Question 8

Do I have to give every developer their own credential?

Accepted Answer

No, but one credential per developer is what gives you attribution per developer. Share one across a team and the attribution is per team — the granularity is a choice, not a constraint.

Question 9

What happens when a developer hits a ceiling?

Accepted Answer

Caps apply per developer, team, feature and model, and whichever is reached first refuses the next call. A refused call reaches no provider and costs nothing, and the owner named on the credential is warned on the way up rather than after the invoice.

Question 10

Does this replace my model gateway or observability stack?

Accepted Answer

Both arrive inside Forgebench, already wired to the credentials, the ceilings and the record, so there's nothing to integrate between them.

Question 11

Do guardrails block or delay a response?

Accepted Answer

No. Input and output are evaluated from the recorded call after it returns, so a check never delays a response or breaks streaming. Every flagged call names the check that fired.

Question 12

Is it self-hosted?

Accepted Answer

Yes, in your own infrastructure, and it reaches models inside your perimeter through the same interface as any provider.

Question 13

Can we govern agents in production too?

Accepted Answer

Yes, on the same deployment. Agent registration, tool authorization and release safety are added on top of the credentials, ceilings, guardrails and record described here.

Question 14

Can Forgebench govern AI usage across multiple model providers?

Accepted Answer

Yes. Forgebench provides a common governance and visibility layer for AI usage across different model providers, allowing organizations to manage usage, attribution and policy through a single control layer rather than managing each provider separately.

Question 15

Can Forgebench work with AI infrastructure we have already built?

Accepted Answer

Yes. Forgebench is designed to work alongside existing AI infrastructure rather than requiring an organization to replace its existing model or AI stack. It can provide the governance, attribution and control layer around the infrastructure already in place.

Question 16

Can Forgebench work with AWS Bedrock?

Accepted Answer

Yes. Forgebench can integrate with deployed models through enterprise AI infrastructure such as AWS Bedrock and provide a unified view of governed AI usage.

Question 17

Can Forgebench work with Azure OpenAI Foundry?

Accepted Answer

Yes. Forgebench can integrate with deployed models through Azure OpenAI Foundry to provide a unified view of AI usage while supporting enterprise deployment and data-privacy requirements.

Question 18

How does Forgebench integrate with our existing SSO and RBAC?

Accepted Answer

Forgebench supports standard SSO and RBAC integrations. Developers can use the organization's existing identity infrastructure, while Forgebench maintains the credentials and usage records needed for AI governance. Agents are registered as distinct identities when separate attribution and controls are required.

Question 19

What happens if an application calls an AI provider directly instead of going through Forgebench?

Accepted Answer

Forgebench can only govern and record traffic that passes through its controlled endpoint. A direct call using an independent provider credential does not appear in the Forgebench ledger. Organizations can compare provider-reported spend with Forgebench-metered spend to identify potential ungoverned AI usage.

Question 20

Can we make Forgebench the mandatory path for AI calls?

Accepted Answer

Yes, when deployed with the appropriate network controls. In a network-enforced deployment, workloads can be configured so that model providers and external endpoints are reachable only through Forgebench. This makes governance an enforced network path rather than an application-level choice.

Question 21

Can we restrict which models a developer or team can use?

Accepted Answer

Yes. Model access can be controlled through the policies associated with the credentials issued to developers, teams or features. This allows organizations to define which approved models can be used in a given context.

Know Every Developer Using AI In Your Org.

There Are Two Ways Developers Use AI In Your Org. Drive Both On One Console.

Credentials

Adoption

Ceilings

Impact

Record

Privacy

Issued From The Console

Split Across Models

Tagged By Owner, Team And Feature

Revocable Live

Integration

Identity & Tenancy

Security

Planning & Setup

Prove It On Your Own Workload. Four Weeks, And Every Phase Exits On Evidence.

Know Every Developer
Using AI In Your Org.

Prove It On Your Own Workload.
Four Weeks, And Every Phase
Exits On Evidence.