How CoderHelm works

CoderHelm has two parts: a gateway that receives events (a labeled issue, a Jira comment, a new push) and serves the dashboard's API, and a worker that does the work. Each piece of work is a message on a queue, so a slow run never blocks anything else.

GitHub / Jira / alerts ──► gateway ──► queue ──► worker ──► agents ──► pull request
                              ▲                     │
          dashboard ──────────┘                     └── your code, code graph, memories,
                                                        skills, data sources, connections

Concept	In one line
Agents	The planner, the coder and the reviewer, each with its own model and tools.
Skills	Your team's instructions in Markdown, read from a git repo.
Memories	What CoderHelm has learned about each repository.
Code graph	A map of where things are defined and who calls them.
Data sources	Read-only access to your databases, APIs, search and logs.
Connections	Outside tools (MCP servers) the agents can use.
Knowledge bases	Documents the agents can search, and a chat that answers from them.
Roles	Who on the team can do what.
Token limits	Monthly caps on new tokens, cache reads and cache writes.
Usage and audit	What CoderHelm did and what it cost.

Everything CoderHelm reads from your systems is read-only unless a page here says otherwise. Code changes always arrive as a pull request you can review.