Runeya

Chat & AI ​

The chat opens with the ✨ (sparkles) icon in the navigation bar, or with Ctrl+Shift+Space. It shows in a panel, next to your services. Conversations are stored on your machine, encrypted.

Two ways to reach a model ​

Through a CLI installed on the machine: Claude Code or Codex. Runeya detects which ones are installed, including on a remote agent, and runs the CLI for you. You do not need an API key: your own installation and subscription are used.

Through an API provider: Anthropic, OpenAI, Mistral, Groq or Ollama. The key is set in Settings → AI providers. You pick the model when you open the conversation.

In both cases, messages go from your machine to the provider. They never go through a Runeya server.

Conversations ​

Answers stream in as they are generated. A conversation can be archived, which removes it from the list without deleting it. It can also be deleted permanently.

All the context in one conversation ​

A conversation's header gives access to the work in progress, without leaving the chat:

  • Git: if the project folder is a git repository, a button shows the number of lines added and removed. It opens the repository state, with changes and commits. These changes cover the whole repository, not only the conversation.
  • Terminal: each conversation has its own terminal, on its agent's machine. It opens in the folder of the open project. The session and its history are kept when you open it again.
  • Kanban card: for a conversation started from a card, the "Kanban" link shows the card's current column, updated live. A click opens the card on top of the chat, without leaving the conversation.

One conversation thus brings together the discussion, the repository state, a terminal and the card for the work in progress.

Starting a conversation from a kanban card ​

With a cloud account, open a board card and click Launch AI conversation. Then pick the scenario to use.

The conversation receives the card's context: the board, the title, the description, the column, the team's instructions, the latest comments and the list of attachments. It can also read the customer conversations linked to the card.

At the end of each answer, the AI adds a comment on the card and can move it to another column. The team follows the progress from the board. From the card, the conversation opens again in one click.

The review companion ​

The companion is a second AI that reviews every turn of a conversation. It reads the actual code in the repository (git diff, changed files and their callers) rather than what the main AI claims. It looks for:

  • what is missing from the request, or what was added off-topic;
  • breaches of the project's rules (CLAUDE.md, AGENTS.md);
  • unhandled errors, regressions and side effects;
  • duplicated code and security issues;
  • weakened or deleted tests, and unchecked claims ("the tests pass").

When it finds a problem, it writes in the conversation, with the file, the line and the severity. The main AI receives that message and can fix it. When it has nothing to say, it writes nothing. You keep a second pair of eyes on the work without reviewing every turn yourself.

In the terminal, the companion does not write while you are typing a message: it waits until you have sent it. Meanwhile, its notes appear in a queue to the right of the terminal. Each one can be dropped (Remove) or sent right away (Send). Send does not wait: if you have text being typed, it goes out with the note.

The companion does not review scenario conversations.

Turning it on for a conversation ​

The Companion panel, under the conversation header, has the Enable the companion button. It also lets you pick its runner (Claude Code or Codex), its model and its effort, for this conversation only. The panel's ⋮ menu offers:

  • Run now: reviews the last exchange without waiting for the next turn to end;
  • Raw transcript: shows the whole exchange with the companion;
  • Test the bridge: checks that its messages reach the conversation.

Its default settings ​

In Settings → Review companion (in the "AI" group):

  • Enable on new conversations: the companion starts with every new conversation. Existing conversations keep their setting.
  • Default companion: the runner, model and effort for new conversations. "Same as the conversation" uses the runner of the conversation being reviewed.
  • Specifics added to the companion: instructions on top of its own, for example "watch migrations and data-loss risks in particular".

These settings belong to the machine. They are not shared with the team.

Attaching images ​

An image helps the model understand an error or a mockup.

  • Claude Code or Codex: paste the image with Ctrl+V in the conversation's terminal.
  • API provider: use the paperclip button above the input box. JPEG, PNG, GIF or WebP formats, 5 MB and 5 images at most per message.

Images are sent to the model like text. Runeya deletes them after 24 hours.

Running a scenario ​

From the AI tab, you start a scenario like a conversation: pick it, name the run and send your request. See Scenarios.

Encryption ​

Messages are encrypted on disk with a key that never leaves your machine. If you lose that key, or the master password that opens it, the conversations can no longer be read.

Runeya