Manage Multiple Agents,Design Cross-Device Loops,Inside the AI Coding Forge.
Run Codex, Claude Code, and OpenCode side by side on real local projects. Diff Forge coordinates terminals, file leases, voice dictation, screenshots, Loop Spaces, and remote visibility.
Ship parallel workCodex, Claude Code, and OpenCode in local terminals
Automate repeat loopsSchedules, scripts, and agent handoffs
Talk through fixesLocal Whisper or cloud voice input
Drop exact contextSnip, annotate, and drag evidence into agents
7.Watch tokens and credits liveReserve and consume events stream by device, workspace, and agent.
Tokenomics
Credit Observatoryreserve, consume, release in real time
reserveconsumerelease
Diff Forge AI Credits
3,840live balance
Usage rate1.8k/min
CodexVoiceClaude
reservevoice session-120
consumetranscript stream-84
releaseunused hold+36
reconciled by device, workspace, and agent
Features
Hire the forge for the whole loop.
Build with parallel agents, command them by voice and screenshot, and observe every token and credit they spend.
repo coordination map
src/authdb/migrateui/sessiontests/e2e
Codexlease held
auth.ts
Clauderisk scan
db.sql
OC
OpenCodeauto wake
ui.tsx
DFlocal kernel
lease ledgerclean
claimsrc/auth/**
reviewmigration order
mergevalidated patch
no collisionsreview gatesnext agent wakes
Parallel Agents, Locally Coordinated
Run Codex, Claude Code, and OpenCode side by side. The local kernel leases files, validates patches, parks conflicts, and wakes the next agent when dependencies clear.
LSloop space
Runway scheduletimed jobs, waits, and agent handoffs
4-step loop
now+1m+2m+3m
00
cron wakes
ready
01
script runs
guarded
02
agent handoff
ready
1wait for lease
2run simple-alert
3send summary
4repeat window
Loop Spaces That Keep Work Moving
Schedule graph workflows that run scripts, send agent handoffs, and keep the next step moving even when a terminal is parked or waiting.
open the failing session testspoken command captured from the bottom bar
speech parser
intentfix session test
modedictate + agent
targetactive app
Local Whisperprivate dictation
Realtime Agentvoice orchestrator
Voice Control and Dictation
Dictate into any app with local Whisper or cloud transcription, then hand larger commands to the realtime voice agent.
context bundlesnip to mark to attach
1
capturearea snip
2
markarrow note
3
attachagent ctx
screenshotviewport evidence
markupwhat changed
promptready to drop
drop payload3 files + annotations copied
Snip, Annotate, Drop Context
Capture screen context with hotkey snips, mark it up, copy it, or drag previews straight into agent terminals.
Mac live
Web lane
VM idle
AI credits3,840
reserve-120
consume-84
release+36
sync manifest
Docsnotes + plans84%
Assetssnips + media62%
Scriptsloop payloads100%
web to MacBookmessage delivered with current cloud state
Cloud Sync, Devices, and Credits
See device presence, sync docs and assets, route messages from the web dashboard, and watch live credit events.
Works with your tools
The agents, speech engines, and storage you already use — wired straight into the forge.
Over a browser tab full of chat windows: real execution, real coordination, and numbers you can trust.
Code stays on your machineMulti-agent terminalsFile leasesPatch validationAuto-wake parked agentsMCPs in every agentDeterministic model defaultsDevice-authoritative numbersBlueprint Loop SpacesScheduled scriptsAgent handoffsRuntime route eventsCode stays on your machineMulti-agent terminalsFile leasesPatch validationAuto-wake parked agentsMCPs in every agentDeterministic model defaultsDevice-authoritative numbersBlueprint Loop SpacesScheduled scriptsAgent handoffsRuntime route events
Install your own agent CLIs, mount MCP servers, and import Skills from the tools you already love. Diff Forge AI connects them to agents, voice, snips, storage, and cloud state.
100,000 credits includedMaximum monthly AI allowance
Up to 3 Dedicated Teams3 teams, each up to 25 members and 50 active team devices, coming soon
BYOC installsRun on your own devices or cloud VMs, coming soon
Automatic updatesCurrent desktop builds
Priority supportFaster plan and setup help
Intensive cloud workBuilt for high-volume AI usage
FAQ
Questions, answered honestly.
What Diff Forge is, what stays on your machine, and what credits actually pay for.
Diff Forge AI is a native Agentic Development Environment (ADE): a desktop app where coding agents like Codex, Claude Code, and OpenCode work on your real repos in parallel, coordinated by a local kernel, with voice control, a snipping studio, and a web dashboard for remote visibility.
Codex, Claude Code, and OpenCode run in embedded terminals. You can also mount your own MCP servers into every agent through the workspace MCP gateway.
Code and execution stay local. The cloud carries live state, commands, and credit events. Audio streams to the cloud only when you choose a cloud speech engine — local Whisper keeps dictation fully on-device.
A bottom-bar recorder gives you push-to-talk dictation in any app, transcribed by local Whisper, your own Deepgram key, or Diff Forge Cloud. A separate GPT-Realtime voice agent can open terminals, queue todos, and drive device controls by voice.
Credits meter Diff Forge Cloud usage — cloud transcription, voice agent sessions, and cloud AI calls — through live reserve and consume events. Your own agent subscriptions and API keys keep billing through their own providers.
Create a free account to start the account path. The Plus plan unlocks the native desktop app license with included credits; Pro and Ultra add larger credit pools for heavier multi-agent and voice work.