Everything you need
to run a Mission.

A Mission is a continuous piece of work with an outcome, context, a plan, execution, and review. Use it for a focused task or a larger outcome that needs several turns.

Install Grain

Download Grain Desktop on every computer you want to connect. Each installation includes the Grain Executor—the background runner that picks up assigned Missions. Sign in with the same workspace, name the computer, choose its root directory, and select which repositories it may use.

Run your first Mission

1. 1
   Describe the outcome

   Try “Review this repository, fix the failing tests, and show me the changes before committing.”

2. 2
   Add context

   Attach a file or choose a project, repository, worktree, knowledge source, or connected app.

3. 3
   Choose where it runs

   Keep Auto-route on, choose hosted Grain Models, or select an Executor and model on one of your connected computers.

4. 4
   Review the plan

   Answer clarifying questions and confirm the plan. Grain keeps the work visible while it executes.

Grain is the orchestration layer between what you want done and the models, tools, files, repositories, and people required to finish it.

Mission lifecycle

A Mission can span several agentic turns. Its conversation, questions, plan, activity, approvals, output, and follow-up edits stay together so you can leave and return without losing the work.

Grain Models

Grain Models run in the hosted Grain environment. They are the fastest path when a Mission does not need a private local repository or a tool installed on one of your machines.

• Available from the web, desktop, iPhone, and iPad.
• Useful for research, planning, writing, analysis, and cloud-connected work.
• Can use attachments only after you explicitly allow the Mission to read them.

Grain Executor

The Grain Executor is the secure background runner included with Grain Desktop. It picks up a Mission assigned to that computer, launches the selected AI tool, works with the local repositories and files you allowed, streams progress to Grain, and returns the result for review. You can start and steer that Mission from the web, desktop, iPhone, or iPad while execution stays on the selected computer.

Your AI Workhorse

Your Workhorse is the fleet formed by your connected Executors, models, and AI subscriptions. Connect Claude Code, Codex, Gemini CLI, Grok, Kimi, OpenCode, and other supported tools, then let Grain choose an eligible computer or route a Mission yourself.

Approvals and review

Approvals stay attached to the Mission. Use them for plan confirmation, decisions that need owner input, and actions with external impact such as publishing, sending, or committing.

Projects give Missions a durable home: a repository, files, instructions, default branch, and the people who can review the work.

Projects and repositories

A project is the ongoing workspace. A Mission is a specific outcome inside—or independent from—that workspace. Repository work can run in the saved checkout or an isolated worktree.

Grain Suite

Grain Suite turns results into working artifacts rather than static chat replies. Open them inside Grain, inspect the source, ask for changes, edit directly, save, and export.

Sheets, Docs, and Tables

Uploading a spreadsheet or document creates visible file context in the Mission. Open it in the matching Grain Suite editor, then continue chatting with the same file attached. “Add a quarterly summary and flag incomplete rows” should operate on that workbook—not create an unrelated page.

Files and context

Attach CSV, XLS, XLSX, documents, PDFs, images, and other supported context from the composer. Before a hosted model can use an attachment, Grain asks you to allow that model to read it.

Git and publishing

For code work, review the diff and verification results before committing. Grain does not assume a repository’s default is main or master; the project base branch is an owner-controlled workspace setting.

Knowledge and integrations let Grain work from the systems your team already uses while keeping access visible and governed.

Knowledge

Connect approved documents, notes, and data sources so Missions can retrieve relevant context at run time. Keep durable workspace instructions separate from one-off attachments.

Integrations

Connect collaboration, product, sales, storage, and developer tools. Each connection should show its owner, permissions, health, and the Missions allowed to use it.

Slack and Teams

Mention Grain where it is installed to start a Mission from a conversation. Follow-up messages in the same thread continue the same Mission. Grain stays quiet unless mentioned or the owner has enabled a specific automation.

MCP servers and tools

Owners can add approved MCP servers and control which tools are available to each workspace or model. External actions remain subject to permissions and approval policies.

Owner controls keep routing explicit across machines and users. They are workspace settings, not hidden assumptions in a local app.

Machines and routing

Each registered machine shows its name, operating system, online state, executor health, available models, and repository access. Pick a preferred machine or allow eligible devices to accept work.

Repository roots

Set a root directory per machine and map the repositories available beneath it. The workspace keeps the logical project mapping while each machine retains its own filesystem path.

Base branches

Owners can choose a default base branch per project. New repository Missions and worktrees start from that branch, and the setting persists across machines and workspace members.

Model health

Send a small readiness prompt—such as “Ready?”—through a device. Grain records whether the model responded, returned an error, or timed out. The owner then decides whether the model should be active for routing.

Security and data

• Local execution reads repository files and local credentials on the selected machine.
• Hosted execution only receives context explicitly provided to that Mission.
• Owners control model, tool, machine, repository, and integration availability.
• Approvals create a visible boundary before sensitive external actions.

Drafts and isolated work

A draft gives an experimental change its own conversation, preview, branch, and worktree. Accept the result when it is right; otherwise keep iterating without touching the main checkout.

Ask AI with your data

Approved data connections let a Mission answer questions from live business context. Grain shows the query or source used so reviewers can verify the result.

Model effort and capabilities

Effort controls adapt to the selected model. Auto-route considers capability, availability, context, and workspace preferences instead of sending unsupported settings to a model.

Git push policy

Repository changes remain reviewable before publication. Workspace owners define the branch and approval policy; a Mission respects the selected project, checkout, and base branch.