Sign in Get Started
Develop anywhere

Deploy Forward

deployment_audit.log
Waiting for response... 0 tok $0.00 0.0s
→ Build Connecting builders
Track Usage scored
○ Deploy Queued
— Builders
On the board
— Tokens
Tracked · status only

We are turning AI tools into adopted, measured deployments. Workflow-first, vendor-neutral, and live implementation in days.

See how it works
Scroll
Deployment Plan
Map what we'll deploy
Fixed-fee scope — you keep the plan
Looking to get your operations AI-enabled.
We'll reply within one business day to scope your deployment.
The Problem

Bills tell you that you spent.
Never why, or who.

Model & harness agnostic
Eight tools, eight blind spots
Claude Code, Codex, Copilot, Gemini, Grok — and the list keeps growing. Each writes its own transcript in its own shape. No vendor totals them, and none of them knows which builder was driving.
Minutes, not month-end
The bill arrives too late
A runaway agent bills for days before an invoice names it. Caps tell you that you hit them; nothing tells you which session, model, or repository got you there.
Per builder, per repository
Waste looks exactly like work
Gross tokens flatter the loudest agent. Until spend is tied to a merged pull request and a named human, you cannot tell throughput from thrash. That is token-maxing.
# Builder Tokens Raw tokens across every tracked harness. Tokens are a denominator, never the ranking — the Board ranks on Build Score.
Reading the board…
How it works

Four steps to attributed spend.

Nothing sits on the wire. The tracker reads what your agents already wrote, and GitHub tells us what shipped.

01
Install
npx --yes deploy-forward@latest on each machine. It reads transcripts your agents already wrote. Nothing else changes.
02
Attribute
Enroll the device in your organization and install the GitHub App. Spend stamps to the org; pushes and pull requests match on the commit login.
03
Fold
Sessions roll into per-builder, per-repository windows. Tokens by model, cost at api-equivalent rates, labeled an estimate every time it is shown.
04
Govern
In the Ledger, set a token or dollar ceiling for the organization, a team, or one builder — then watch it fill as work lands, in minutes rather than at month-end. 80% reads warn, 100% reads over, and a burn 6× the team median flags itself without anyone watching.
Step 02 is where teams stall. A 30-minute onboarding session gets your org connected and reconciled against your own invoices. Book onboarding →
Open Source

The capture layer is open.
Read it before you trust it.

The code that reads a transcript, counts a token and prices a model is MIT licensed and public. Run it, audit it, fork it. A number you cannot check is a number you should not report to your CFO. The tracker is open and the Board is public; the Ledger is neither, because it is your organization’s spend.

$npx --yes deploy-forward@latest
View the repository
What is open
The tracker: every harness adapter, the hooks integration, the pricing table and the token math. MIT, shipped with the package.
Claude CodeCodex Copilot CLIGrok OpenClawopencode HermespiGemini and growing
The Product

The Board, and the Ledger.

One tracker feeds both. The Board is how your engineers show what they built. The Ledger is how your company knows what it cost.

The Board
— builders · public
The Ledger
Behind auth · your org only
Source
One tracker. MIT, running on the builder’s machine, reading transcripts your agents already wrote. Everything below is a fold of what it sends.
Answers
Who built what, and what it took.
What it cost, and who spent it.
Grain
One builder, ranked against the rest.
Member, repository, project.
Visibility
Public. Anyone with the link.
Your organization. Nobody else.
Budgets
None. It reports, it never governs.
Warns at 80%, escalates at 100%. Never blocks: every dollar here is an estimate, and an estimate should not stop a build.
Price
Free. No account needed to read it.
Paid, per organization.
Open the board
Connect an organization
The Model

Embed. Match. Deploy.

We embedded inside real operations to learn how AI actually gets deployed, rather than advising on it from the outside. Everything below is what that taught us. The tracker exists because we needed to know what the work was costing while we did it.

STAGE 01
Embed

We sit inside the operation and watch the work move: which teams, which systems, which handoffs, and where the hours actually go. Nothing gets recommended before it has been observed.

STAGE 02
Match

Then we pick the path that fits the workflow: a vendor tool, a partner product, a custom build, or one of ours. Vendor-neutral by design, and often the answer is that a workflow is not ready for an agent yet.

STAGE 03
Deploy

We integrate it into the workflow, train the team on it, and stay until it holds without us. Then the same measurement runs on it that runs on us: spend against deployments, per team, per repository.

How a deployment runs, start to steady state
Embed
Match
Deploy
Train
Scale
Spend is tracked at every stage, not measured at the end. The line through those five is the tracker, running from the first day of the audit.
How we run an engagement
The Solutions

Built to deploy. 
Developing what's next.

What we have built for other operations, and how an engagement actually runs, is on the Solutions page.

How engagements work →