Collect
Connectors read messages, issues, pull requests and notes into a local event log.
Private work agent for overloaded teams
Echelon reads the work streams you already live in, groups related activity into durable cards, drafts the next move and waits for your approval before anything leaves your machine.
Three threads and one pull request describe the same issue. Echelon keeps the context together and suggests the next reply.
Client asked for a deployment date. No message will be sent until you choose one.
Two owners found, one follow-up drafted, no noise added to the board.
How it works
Connectors read messages, issues, pull requests and notes into a local event log.
Related conversations become one work card, with the owner’s replies kept in the timeline.
The agent drafts replies, next steps and handovers with citations to the source material.
The queue asks for approval. Sending, rejecting and changing policy stay human decisions.
Echelon is local-first. Credentials stay in the OS keychain. Drafts queue for approval. Automation is scoped by policy, per platform and per recipient.
No cloud telemetry by default.
Every outbound message passes through a send gate.
Model providers are modular: Claude Code, API keys, local models or ChatGPT Codex.
Now in private build