04 / AI outreach
Eon.
Automation gets dangerous when it is allowed to speak for you. Eon prepares the research and the message, then stops and asks a person to approve.
- approval before send
- Human Recorded 2026-10-31
- campaigns
- Budgeted Recorded 2026-10-31
- self-hosted build
- Private Recorded 2026-10-31
- send history
- Recorded Recorded 2026-10-31
- 01
Research
A sourced profile and a reason to write, not an invented personal detail.
- 02
Approval
A person reads the draft, edits it, or decides not to send.
- 03
Send
The campaign budget applies and the action is written to an accountability record.
01 / Context
The problem
Outreach tools make it easy to send more. That is often the wrong goal. A generic message sent at volume damages the sender's name, and an autonomous agent that writes without a check can turn a small list into a liability. The useful problem is not volume. It is helping a small team do careful outreach without losing the judgment that makes it worth reading.
Eon was built as a private system for that problem: research first, a draft next, and a person in the last position before anything leaves.
02 / Audience
Who it is for
A small team that needs to contact the right people and cannot spend a day assembling context for each message. The work still belongs to a person. Eon removes the repetitive collection and formatting around the decision, not the decision.
03 / Constraint
The constraint that shaped the build
The system has to know where a claim came from. A profile assembled from the wrong person or an invented fact is worse than no profile. The approval step has to be fast enough to use, and the budget has to stop a campaign before it becomes an accident.
The build is self-hosted, so the operator controls the data and the sending path. That choice adds operational work: deployment, storage, and delivery have to be part of the product, not an afterthought.
04 / Build
What we built
Eon is a chain of small, inspectable steps:
- A research queue that assembles a sourced profile and the context for a message.
- A draft and an approval screen, where a person can edit, reject, or send.
- Campaign budgets that constrain the work before it reaches the network.
- A record of what was sent, to whom, and which approval allowed it.
- A self-hosted deployment with a Docker and Linux base.
05 / Decisions
The hard part
The interesting constraint is not writing a good sentence. It is keeping the evidence attached to the sentence and making the approval state impossible to skip. A queue that can be bypassed is not a control. A budget that lives only in a spreadsheet is not a budget.
The system also has to stay useful on a small list. If the research step takes longer than reading the same pages by hand, the tool is not doing its job.
06 / Outcome
Why there is no public link
Eon is a private build. There is no public demo, and this page does not pretend there is one. The case study is here because the design decision is useful: human approval, a budget, and an accountable record can be part of an AI workflow from the start.