Connect your AI agent to the tools you already use and run outbound built for your market, your offer and your process. Campaigns, specialist roles, approvals and memory live on your machine.
The shift
GTM no longer starts with buying a tool per job. With an LLM, any founder or operator can build the exact workflow their business needs.
Then
A stack of SaaS: one for data, one for sequences, one for enrichment, one for the inbox. Each with seats, a UI and its own idea of your sales process.
Now
One AI agent can research a market, write to a person and call any API. You describe the workflow in plain words, and your agent builds it.
The gap
When the chat ends, nothing remembers what was sent, what waits for a decision or what happens next. Outbound runs for weeks. Jason is the engine for that line.
You describe the process. Your agent builds it with skills and roles you can read and change.
Runs on your machine with the agent you already use. No Jason model, no telemetry.
A provider-independent core with a public plugin model. Reply ships ready to use; other tools connect through plugins.
Who it’s for
Founders
GTM operators
Lead gen agencies
How it works
Four parts, one job each. Only the runtime is Jason’s; the agent and the provider accounts stay yours.
thinks
The AI you already use. Woken only when judgment is needed.
teach the trade
SDR expertise carried by nine roles, from planner to reply handler.
keeps it going
Durable state on your machine. Survives restarts and crashes.
reach your tools
Reply and ntfy ship today. Others connect via public contracts.
You
Tell your agent the goal in your own words.
Planner
Turns it into an ICP, segments, channels and a first small test.
Researcher · Copywriter · Critic
Build the audience, write, personalize and review.
Approval queue
Everything that would reach a person waits for your yes.
Reply plugin
Approved work goes out through Reply.
Responder · Analyst · Manager
Handle replies, read results and replan when evidence changes.
Anything that reaches a person waits for your approval, by contract.
Your AI on its own
An agent calling a sales API directly
Campaigns that outlive the chat
Lives inside one chat. Close it or restart, and the campaign is gone.
AI decides, code executes
Every send, check and inbox poll goes through the model: tokens every time, a different result every time.
Cold-outreach know-how
A smart generalist. The playbook is whatever you typed into the prompt.
A line for every person
One static sequence for everyone, or one-off messages with no plan behind them.
You step in where it counts
“Ask me first” is a line in a prompt. One wrong tool call lands in a real inbox.
Watches the inbox and comes to you
Sees replies only when you remember to ask.
Holds up on edge cases
After a timeout it guesses. Retry and risk a double send, or skip and lose one.
Your AI + Jason
The same agent, inside the GTM harness
Runtime · durable stateCampaigns, contacts and pending work are kept on your machine and keep running through restarts and crashes.
Dispatcher · pluginsThe mechanics run as plain code: same input, same call, zero tokens. The model is woken only when judgment is needed.
Skills · nine rolesSDR expertise as skills that keep improving, carried by nine specialist roles.
Planner · per-contact workEvery touch is planned for the person it reaches and revised as replies, bounces and silence come in.
Approval gatesA send waits for your yes on the exact text. A notification tells you when something waits.
Inbound sync · managerReplies are read by code on a schedule. The campaign manager looks at what moved and reacts.
State machine · journalA send that may already have gone out is never blindly repeated, and an append-only history records what ran.
Install
One line in a terminal, or one prompt pasted into your agent. Your agent does the rest and checks that everything works.
Install Jason AI GTM Harness from https://jasonharness.ai and check that everything is ready. [Final prompt text TBD]Apple-silicon Mac, Windows x64 or Linux x64. Claude Code installed and signed in.
Paste the prompt or run the line above.
It shows what’s ready, optional and missing.
Sign in through the browser — no secrets in the chat. Or stay standalone for now.
“Book demos with heads of growth at DTC brands in Germany.”
Approve, or reject with a comment that goes back to the roles.
Compare
DIY gives full control, but you build memory, approvals and scheduling yourself. A closed AI SDR is finished, on the vendor’s terms. Jason gives you the finished engine and keeps your AI and your workflow.
| DIY: your LLM + an outbound API | Closed AI SDR SaaS | Jason AI GTM Harness | |
|---|---|---|---|
| Who does the thinking | Your LLM, in one chat | The vendor’s AI | Your own agent, split into nine specialist roles |
| Memory between sessions | None; you rebuild context each chat | Kept in the vendor’s cloud | Durable runtime on your machine |
| Fit to your workflow | Whatever you script | The vendor’s playbook | Your process, through skills and custom roles |
| Approvals | You build them | Set by the vendor | Every action that reaches a person waits for you |
| Follow-ups and reviews | You build the scheduler | Handled by the vendor | Dispatcher plus a manager role on events and cadence |
| Reply handling | You build it | Handled by the vendor | Responder classifies replies and drafts the next move |
| Tool choice | Any API you wire by hand | The vendor’s stack | Provider-independent core; Reply ready to use |
| Where your data lives | Wherever your scripts write | The vendor’s cloud | Jason’s state stays local |
| Cost model | LLM + API usage + build time | Vendor subscription | Free runtime; Reply credits for what you send |
| Setup | Days to weeks of building | Vendor onboarding | One-line install, guided by your agent |
| Best for | Engineers who want to build everything | Teams who want done-for-you | Operators who want their workflow without the plumbing |
Closed AI SDR examples: 11x, Artisan, AiSDR. Vendor details based on public sites; [source and date TBD]. We make no claim of better results.
Pricing
Start with 1 mailbox and 100 sends. Then prepaid Reply credit packs. No seats, no subscription.
Free runtime
1 mailbox + 100 sends included
Prepaid Reply credits after that
No seats, no subscription
FAQ & honest notes
No. Jason is in alpha (v0.9). Expect rough edges and frequent updates.
It’s designed for compatible agents. Today, background roles run in Claude Code.
Reply and ntfy plugins ship. Any other tool needs a plugin written against the published contracts — your agent can help write it.
Windows x64, Linux x64 and Apple-silicon macOS. Intel Macs aren’t supported.
Skills, plugins, contracts, docs and install scripts are MIT-licensed. The runtime is a free executable whose source isn’t published.
Running Jason as an outsourced service for other companies needs a written agreement. Get in touch and we’ll walk you through it.
We don’t promise replies, meetings or revenue. Jason runs the operation; results depend on your market, offer and process.
No Jason model, no telemetry. Jason’s state stays on your machine; your AI and providers get only what they need.
Install in one line and run your first campaign in your own words.