← newsletter

DevLog #15: The July Roundup

From Login to Your First PR, in the Browser

Getting started no longer requires a raw cloud shell. A first-run org is guided through setup and can reach a real pull request without leaving the app.

• Set up your whole org from the browser. A guided, card-based hub walks a new org through identity, providers, chat, code access, a dev environment, and integrations, with an embedded terminal login and an in-browser agent to help. The whole setup happens in the browser.
• Go from an empty account to a real pull request. A guided track connects GitHub, lets you pick a repository (by URL if the list does not load), and has the agent open an actual pull request.
• Turn a repeated task into a skill. Save a recurring team workflow as a reusable skill that your agents load automatically in every session.

Durable, Shareable Conversations

This was the most important structural change of the month. A conversation is now stored independently of the runtime executing it, which enables a few things you can do for the first time.

• Close a session and come back to it. Resume a closed session later on a fresh cloud machine with its history, title, and order intact. The history sidebar shows one row per conversation, with no duplicates after a reconnect.
• Work in the same session as your teammates. Several people, tabs, and surfaces can attach to one session and watch the agent’s turns stream live and in sync, so everyone follows the same conversation as it happens.
• Watch a web app come up as the agent builds it. A preview drawer beside the chat renders the web app your session is running, such as a dev server or a generated page, and updates as work happens. It has a full-screen mode on mobile.

More Control Over Scheduled Work

Triggers run Nori’s unattended work, such as cron jobs, webhooks, and scheduled maintenance. In July you gained real control over how they run.

• Choose which agent runs a trigger. Pin a specific coding agent (Claude, Codex, Gemini, or Cursor) per scheduled or webhook trigger, from a dropdown or the CLI. It falls back to the fleet default when unset.
• Give a trigger admin access, with approval. Triggers can run with admin scope to reach admin-only secrets. Admins grant this directly, and other users’ requests are staged for approval.
• See when a trigger fails. When a trigger started from a Slack or Discord thread fails, Nori posts the failure back into that same channel so you notice it.

New Integrations

• Set up Slack yourself. Orgs can install and manage their own Slack app during onboarding, either on a shared Nori-hosted workspace or a dedicated per-org app.
• Call the right agent by name. Define named agents in your workspace, each with its own provider, model, and skillset, and @mention one in a thread, for example a reviewer or a researcher. Threads show a live status while the agent works.
• Connect Luma in one step. Luma is now a managed integration. Connect it and Nori distributes a Luma API key to your sessions and installs the nori-luma CLI, so an agent can work with your events.
• Hand a session a secret without exposing it. When a session needs an API key, token, or credential file, Nori serves a secure one-time form. The value is never logged or persisted.

Run Agents Without the UI

• Run an agent from a script or CI. nori exec "prompt", or piped stdin, runs an agent as a one-off command and prints the answer to stdout. You can also expose Nori as an ACP-over-stdio agent for other programs to drive. A bypass flag covers trusted unattended runs.
• Drive cloud sessions from your terminal. nori cloud lists your existing cloud sessions before claiming a machine, and supports the full lifecycle: reattach to a running session, /close to release it, or quit to leave it running.

More in the CLI

/fork can branch a conversation at its current point while keeping the original resumable, /compact runs in place on agents that support it, /browser lets you pick a throwaway, persistent, or system Chrome profile, and the prompt composer gained Vim editing.