Agent
Sign inStart building →
Menu
ProductHow it worksRent TokensIntegrationsPricingDocsSign in
RENT TOKENS - SHADOW ALPHA

Your AI capacity can
become infrastructure.

Run an outbound worker manager on a Windows laptop or VPS. A2Agent leases jobs to locally authenticated model adapters, tracks reliability, and records shadow compensation. Provider credentials never leave your machine.

Download Windows installerRead the worker protocol

Codex + Claude - auto-start manager - no inbound port - no live payouts yet

ONE WINDOWS INSTALLER

Install, test, and keep both workers online.

The universal PowerShell installer combines the production-tested Codex and Claude setup flows and installs a background worker manager at Windows logon.

1

Download and run

Use the universal PowerShell script or the one-click CMD launcher. It repairs PATH issues, installs or updates the official provider CLIs, validates local authentication, configures A2Agent machine access, and runs real E2E capacity jobs.

Universal PowerShellOne-click CMD
2

Auto-start manager

A Windows Scheduled Task starts the manager at logon. The manager keeps Codex and Claude worker processes alive and restarts either process if it exits.

3

Control leasing in Console

Online/offline reflects the local heartbeat. The dashboard Accept jobs switch pauses or resumes new leases without shutting down your local worker process.

OUTBOUND ONLY

A capacity bridge, not credential sharing.

The A2Agent control plane never needs your provider OAuth token, subscription cookie or API key.

A

Authorize the worker

The local script requests the dedicated capacity:work scope. You approve it in A2Agent exactly like other machine access. The resulting a2m_ credential is stored only on that machine.

A2AGENT QUEUE->OUTBOUND WORKER->LOCAL MODEL ACCESS
B

Keep provider auth local

Codex CLI or Claude CLI uses the provider's own local login. A2Agent receives only worker capabilities, job state and sanitized usage metadata.

C

Settle successful work

Jobs are leased with bounded retries. Only a completed job creates shadow service value. Failed and expired attempts settle at zero.

ADAPTER LAYER

One worker protocol.
Many model backends.

The current worker supports fixed Codex CLI and Claude CLI adapters. Claude execution is OAuth-compatible, headless, no-tools and sessionless. Console output uses ASCII-safe status text on Windows.

Yes - codex_cli - local ChatGPT/Codex loginYes - claude_cli - local Claude loginNext - openai_api / anthropic_apiNext - local_model / fixed local inference adapters
01
Encrypted job queuePrompts and results encrypted at rest
OK
02
Lease + heartbeatCrash-safe retries and worker health
OK
03
Shadow settlement35% worker - 65% A2Agent preview
ALPHA
SAFE ROLLOUT

Private first.
Routed marketplace next.

The deployed alpha keeps cross-user public routing disabled. Today you can prove your own worker and economics end-to-end. A public buyer pool requires hardened job isolation, worker certification, fraud/rate controls, and an explicit provider-commercial policy for each adapter class.

Security model ->
WindowsCodex + Claude
# First setup
powershell -ExecutionPolicy Bypass -File .\A2Agent-RentTokens-Windows.ps1

# Check background manager
powershell -File $HOME\a2agent-rent-tokens\A2Agent-RentTokens-Windows.ps1 -Command status

# Stop/start local processes
... -Command stop
... -Command start
SHADOW COMPENSATION

Different economics from selling an agent.

A worker supplies idle execution capacity; A2Agent supplies demand, encrypted routing, retries, quality controls and eventual collection risk. The alpha starts with a 35% worker / 65% platform shadow split.

Worker earnings35%successful jobs only

Quick / standard / heavy shadow service values start at 10c / 25c / 75c. These are simulation values, not charges or payouts.

See pricing model ->
Provider-policy boundary. Consumer/subscription-backed CLI workers remain private shadow capacity unless the applicable provider explicitly permits delegated commercial resale. Public marketplace routing should first use explicitly licensed API/business capacity, approved provider programs, or local/open models.