← newsletter

DevLog #23: Resume Released Sessions, Watch the Agent’s Browser, and Chain Automations

Resume a released session with its files or its conversation

• Get back the work in every repository when you resume. Each checkpoint records every git repository in the workspace, including nested repositories and worktrees: the current commit, staged and unstaged changes, and local branches. Resume restores them onto a fresh machine. Ignored files, stashes, and files outside a repository are not saved, and the base commits must still exist on the remote.
• Resume a session that has no checkpoint. The agent receives the recent conversation and a note that files, branches, and uncommitted changes were not restored. In web chat, a note under the Resume session button says whether resuming brings back the conversation and files, or only the conversation. Replying in a released Slack or Discord thread, or sending !resume, resumes it automatically.
• See when a session is restoring. In Slack and Discord, a reply that wakes a session gets ⏳ until the agent is ready, then 👀. Web chat shows a Queued message and, after 10 seconds, “Restoring your workspace from its last checkpoint…”.

Watch and take over the agent’s browser

• Watch the agent use a real Chrome window, live. Each session has one visible Chrome that the agent controls with the nori-browser command. Open it with the Browser button in the web chat header, or from the Browser: link in !urls in Slack and Discord. The view requires your Nori sign-in.
• Take over to sign in, enter a 2FA code, or solve a CAPTCHA. The agent can stop and ask you to finish a step, then continue in the same browser. The live browser is available on sessions that run on Modal, the default machine provider for hosted Nori.

Chain automations and keep them running through restarts

• Start an automation when another one finishes. Choose “After another automation” when you create one, then pick the source automation and whether to run on any finish, on success, or on failure. From the CLI, use nori-sessions triggers create --type internal-event with --source-trigger-id and --source-event. An automation cannot use itself as its source.
• Count on scheduled, reminder, and webhook automations to run. Each firing is saved to a queue before it starts. A firing that fails to start is tried up to three times, and firings missed in the hour before a restart run after it. A webhook returns 429 when 10 firings for that automation are already waiting.
• Choose what an automation does when every machine is busy. “Reclaim an idle runtime if capacity is full” ends the session that has been idle longest (at least 10 minutes) and posts a notice in its thread. “Wait for a free runtime” holds the firing until a machine is free.
• Manage automations from one page. The Automations page shows each automation as a card, with one side panel to view, create, edit, review, and run it.

Connect API providers and several subscription accounts

• Run your agents on models from OpenRouter, Fireworks AI, Subconscious, Cloudflare Workers AI, or any compatible API. Add a connection under Integrations, pick models from the discovered list or type a model ID, and set a default model for new sessions. Connections work with Claude Code, Codex, OpenCode, and pi where the API is compatible.
• Save several Claude and ChatGPT subscription accounts. Use Add account on the provider card and choose which account new sessions use. Running sessions stay on the account they started with. Members can view accounts, and admins can manage them.
• See how much subscription allowance is left. Each account card shows 5-hour and weekly usage bars with reset times, measured from your sessions’ activity.

Run OpenCode and pi, and use new default models

• Choose OpenCode or pi as the agent for a session. Connect either one to an OpenAI-compatible endpoint with a base URL, an API key, and a model.
• Start new sessions on Claude Opus 5.5 or GPT-6 Sol by default. Opus 5.5 is the default Claude model and GPT-6 Sol is the default ChatGPT model for organizations that have not chosen their own default.
• Choose Claude Fable 5.1. Select it in the web model picker, with !model in Slack, or in automations.
• Move Gemini CLI work to another agent. Gemini CLI has been removed as an agent. A thread, automation, or profile that still names Gemini stops with a message asking you to pick another agent.

Have Slack agents reply only when tagged

• Set agents to reply only when someone @mentions them. Choose Tag only under Security, then Slack replies, or run nori-sessions fleet set-setup --slack-response-mode mentions. Untagged messages are saved as context for the next mention, and ! commands still work. Custom agents can have their own setting, and !tagonly on, off, or default sets it for one thread.
• See a session’s setup at the top of each new thread. New Slack and Discord threads start with a card showing the provider, model, skillset, agent, reasoning effort, and a transcript link. Admins choose which fields appear under Settings, Session environment, Thread startup card.

Work in the redesigned web chat

• Read each conversation as a series of turns. The agent’s tool calls fold into one line above its reply, and an outline in the margin lets you jump between turns. Pick the model, provider, and skillset from searchable menus before your first message.
• Attach files and images to a prompt. Use the paperclip button, drag files in, or paste. Files up to 25 MB are uploaded to the session machine, and images are also sent to the model.
• Run commands from a slash menu. Type / to run session commands, switch model, provider, or skillset, and insert skills from your active skillsets.
• Organize conversations and use Nori on your phone. Rename, pin, archive, delete, or release a conversation from the sidebar, and group or filter by date, source, or status. Install the chat as an app on your phone or desktop, and switch between light and dark themes.