What Sasha is and what Sasha holds
Sasha by Context is Everything is a private AI knowledge base. Each organisation has its own instance, which no other organisation shares.
Every page of the Sasha docs, with the tab it is in and its kind.
Sasha by Context is Everything is a private AI knowledge base. Each organisation has its own instance, which no other organisation shares.
Connect Claude to your organisation's Sasha so you can search your knowledge and run Sasha recipes from Claude, never with more access than you have in Sasha.
Sasha keeps your documents in projects. See what a project and a member's home are, and which projects each person can reach.
Each person in Sasha is an admin, staff or member. See what each role can do, what an admin can change, and what a Claude connection can reach.
Ask Sasha a question about a rule in your own policies and get a short answer, the exact wording it comes from, and a warning when two documents disagree.
A chat lives with the agent or project open when you start it. See where a new chat goes before you type, how the left menu lists chats, and how to move or delete one.
Ask Sasha to check what your documents say about one topic and get the places where they disagree, the questions they leave open and what may be out of date.
Ask Sasha what your organisation already knows about a topic and get a short summary, with every claim linked to the document it came from.
In the Sasha web chat, a card shows when one step of a reply runs for more than 30 seconds. See what the card says and what to do.
Ask Sasha for a short brief before you meet a client or prospect, built from your past meetings, open issues and files, with every source named.
Ask Sasha to summarise a saved meeting record or transcript and get each decision with who made it, and each action with its owner and due date.
An agent writes drafts but never sends them. Read each draft, then send it, change it first, mark what you did yourself, or bin it.
An agent does one regular job for your business, such as following up quiet customers or checking your costs, and holds its work for you to approve.
A schedule runs an agent or a skill at set times, for example every Friday at 4pm or on the 2nd of each month, so that a regular job happens without anyone asking.
Prepare the sources, start the set-up from Agents that work for me, check the brief, choose what the agent may read, run the first report, then put it on a schedule.
Connect one folder of a GitHub repository to a Sasha project. Sasha keeps a read-only copy and refreshes it every day. Make a read-only token, add the card, and keep it working.
Price a service, package, product, course or membership with Sasha, starting from your own pricing documents, with a fairness check and a way to test it.
An app is a small application built on your knowledge, inside Sasha. See who can use and change an app, and how to use apps from Claude.
Add a service's MCP server in Sasha Settings, keep credentials out of chat, and use safe diagnostics to save, test and discover its capabilities.
Send enquiries and content calendar entries from a bernard website to Notion boards as cards, and let Done or Skipped in Notion update bernard.
A skill is a packaged way of doing a task that Sasha can run. See who can run and build skills, and which skills Claude can run.
When Sasha gets something wrong or cannot do what you ask, it tells the Sasha team in your words and may ask what you expected. See what is sent and why.
What to check and change when a tool connected to Sasha, or Claude Desktop, does not do what you expect, in plain steps.
Settings → AI Admin connects Sasha to an AI provider (Anthropic, a Claude subscription or AWS Bedrock), chooses the model and sets the context window.
The steps in the AWS console that give you the Bedrock API key and region that the AWS Bedrock card in Settings → AI Admin needs, and the settings that stop Fable models failing.
Settings → Brain Update brings the latest shared prompts, guides, skills, agents and system documents from a central repository into your Sasha.
Settings → Rules holds the instructions Sasha follows in every chat (Global rules), the instructions copied into each new project (Project rules), and, for admins, response Styles.
Settings → System shows a health score, CPU, memory and disk use, chat activity and service state, and lets you pause or run scheduled jobs.
Settings → Users is where an admin adds people, sets roles (admin, staff, member), grants members shared projects, resets passwords and 2FA, and removes people.
Settings → Version shows the version your Sasha runs and the newest release, and gives admins three maintenance actions for skills, docs and chat folders.
Settings → Connections adds your Sasha to Claude, makes tokens with the access you choose for Claude Desktop, Claude Code and scripts, and shows 7 or 30 days of usage.
When no Sasha tool can make a change, Sasha prepares a job for Claude Desktop. You approve it in the chat, Claude does it in your browser, and Sasha checks it.
Settings → Tools lists the tools Sasha's AI can use, adds your own MCP servers (with keys or with a sign-in, such as Bernard) and sets which tools a chat may use.
OneDrive, SharePoint, Google Drive, Notion and GitHub synced folders side by side - where each shows in Sasha, how fresh it is, what read-only means and how Claude reaches it.
The GitHub synced folders section of Settings → Shared Drives - the form, the card, the grey folder in the project, the daily refresh, and what to do when a refresh fails.
The steps in Google Cloud Console that give you the client ID and client secret that a Google Drive connection in Settings → Shared Drives needs.
The HireBest card in Settings → Connections links one person's HireBest account to Sasha, so new chat messages can use that person's HireBest projects and research.
The steps in Microsoft Entra that give you the client ID, secret value and directory ID that a OneDrive or SharePoint connection in Settings → Shared Drives needs.
The steps in Notion and in Settings → Shared Drives that put your Notion pages in Sasha's file tree, and how to fix an empty Notion folder.
Set up website events once in chat, so bernard form enquiries and calendar entries become Notion cards, and Done or Skipped on the board updates bernard.
Settings → Shared Drives signs Sasha into one business OneDrive, SharePoint library or Google Drive, so people can browse those files in Sasha without uploading them.
The Files panel shows every folder and document in your Sasha. See how uploads become Markdown, how previews and downloads work, and why member-home files open isolated.
Sasha finds documents by reading them at search time, with plain text matching. See what Claude searches, what Search Chats searches, and why results differ.
The Projects list in the side menu holds your projects, folders and the Member homes folder. See what create, rename, archive and delete change, and who sees which project.
The chat screen is where staff ask Sasha questions in a project. It covers Design & Plan Mode, attachments, stopping a run, question cards, saved files and Quick Settings.
The Skill Library lists every skill. Tasks adds and deletes schedules in any time zone and shows failed runs. Another system can start a skill by API key.
Menu → Agents lists Sasha's specialists and lets you read and edit them. Today no control makes a specialist active in chat, and guides are not applied.
The editor shows a skill or workflow as a diagram next to its text, with inspectors to change steps and settings. Gates and approvals are not enforced at run time.
Apps are small applications inside Sasha that keep their own records. Open them from Agents that work for me in the side menu, use them from Claude, and know their limits.
A Finance Agent or a Marketing Agent is an agent that Sasha's Agent Creator builds for your business. Set it up in the side menu, read its report and email, and correct its brief.
The Meeting Room sends a bot into a Teams or Google Meet call. It shows live captions and advisor tips, and saves the transcript in the project.
Where Sasha keeps files and chats, how long chats are kept, what your AI provider keeps, what data leaves your Sasha, and how Sasha protects the secrets it stores.
A simple explanation of Sasha permissions. Role, projects and connection, shown through the people of one small advisory firm and the jobs they do.
How Sasha's permission model works behind the three roles, what an API key from My Account can do, what service principals are, and which audit records Sasha keeps.
How sign-in works in Sasha, what locks an account and for how long, how to reset a password, how to set up and reset 2FA, and how long a sign-in lasts.
How Sasha files improvement suggestions from chats and from Claude, what the team that improves Sasha receives, what each status means, and how Report Issue differs.
Two ways to publish from Sasha. Report Publisher makes decks and documents from chat, and Bernard edits a website.
What works today when Sasha hands a job to Claude Desktop, what is planned next, and what to do until then.
Ask Claude Desktop to find the job you approved in Sasha, do the clicks in your own browser, and send its report back to Sasha. You copy nothing.
When Claude is connected to Sasha, Claude can search and read your knowledge, run recipes and skills, use apps and, if your Sasha allows it, write documents.
When a job needs clicks in a web app that Sasha's tools cannot do, Sasha writes the steps and Claude Desktop does them in your own browser.
Each organisation has its own Sasha instance. See what Sasha keeps in your instance, who can reach it, and how a Claude connection reaches it.
The reference for agents that use Sasha through MCP. How to connect, how scopes and roles set the tool list, and one page for each tool and tool family.
The changes to the Sasha MCP server and this reference, newest first, so an agent or a developer can see what is new since they last looked.
Every scope of the Sasha MCP server, the consent text a person sees, and the tools each scope opens.
Find a fact in the knowledge base, read the document that holds it, and give the person the fact with the path of its source document.
Start one of the organisation's skills, poll checkExecution until the run ends, and read the document that the skill saved.
Report a bug, a friction, a docs gap or a feature you need to the Sasha team, then check in a later session what happened to it.
Read rows from a Sasha App, total them, add a record through the app's own action, and correct one row, with a person's approval for each change.
Find the brains a connection can see, learn what each kind should hold, find the gaps, and answer from a brain and its parent brains.
Create a note in a member's own home with writeDoc, then change one passage of it with editDoc, with a version check on each change.
Every tool of the Sasha MCP server in one table: what it does, the scopes it needs, and whether it only reads.
Every tool and tool family of the Sasha MCP server, grouped by the job it does, with one line on what each one is for.
List the projects this connection can see. Call it first, then pass a project id to listDocs, searchKnowledge or readDoc.
List the documents in one project, with titles from frontmatter. Paths have no project prefix, so pass the same project to readDoc.
Find documents that contain a text, across every visible project or one. Returns paths with the project prefix, line numbers and snippets.
Read one document, in pages of up to 50,000 bytes. Give project and path, or one path that starts with the project id.
List the brains this connection can see, with each brain's blueprint, level, readable parent and depth by ring. Then open one with getBrain.
Show one brain's contents page - its card, depth by ring, each document with its state and a readDoc path, and the documents it drops.
List the blueprints, or show one - the plan for a kind of brain, with every document's file, title, ring, spine domains, out-of-date rule and asks.
Search the documents of one brain, and with includeParents its readable parent brains. Returns the brain, document title, path, line and snippet.
Read the usage guide on this instance. Call it with no arguments for the table of contents, then once for each section you need.
Tell the Sasha team about a bug, a friction, a docs gap or a feature you need. One call for each issue. Returns an id and a FEEDBACK reference.
Read the status of an improvement suggestion by the id or the FEEDBACK reference that suggestImprovement returned. A member sees only its own connection's suggestions.
List the improvement suggestions filed on this instance, newest first, with each FEEDBACK reference and what happened to it. Filter and page.
Poll a skill run that a skill_ tool started. Returns its status, a summary when it ends, and the path of the document it saved.
Read the transcript segments of a live or recent meeting by its id. For a finished meeting, read the saved document in the knowledge base.
List every source file in a Sasha App package you manage, with its size and version, and the package status. Start every app change here.
Read the full text of one file in a Sasha App package you manage, with its version. Pass that version to the tool that changes the file.
Create a new file in a Sasha App package, or replace a whole file. The change is live at once, with no draft and no undo, and the result says if the app still validates.
Replace one exact passage in a Sasha App source file. The passage must occur exactly once. Live at once, with no undo; the result says if the app still validates.
Delete one file from a Sasha App package you manage. Never APP.md. Remove every reference to the file first. Live at once, with no undo.
Validate a whole Sasha App package you manage again. Says whether the app is live, or lists each problem by file, line, code and reason.
List every schedule on this Sasha: its skill, when it runs and in which time zone, whether it is on, its last and next run, and the last error.
Add a schedule that runs a skill at set times in a time zone. One skill can have several schedules, one for each time it must run.
Change a schedule: pause or restart it, or change its skill, time, time zone or time limit. Send only the fields that change.
Delete a schedule. The skill itself is not changed. To stop a schedule for a time and keep it, pause it with updateSchedule instead.
Start one run of a schedule now, to test it. It returns at once; call listSchedules after a few minutes for the result.
Sync a folder that Sasha copies from GitHub now, and see the commit, the time of the last good sync and any error. Admin only.
Prepare a job for Claude Desktop to do in the person's own browser. The person approves the whole text in a Confirm box before it is saved.
Read a Claude Desktop job's status, the questions Claude Desktop asked and its report, from the chat that prepared the job.
Record Sasha's own check of a reported Claude Desktop job, checked or failed with a short summary, from the chat that prepared the job.
List the Claude Desktop jobs the person approved in Sasha that are not finished yet: ready, claimed or needs_you, newest first, at most 20.
Read the whole text of one approved Claude Desktop job, exactly as the person approved it, and start it: a ready job becomes claimed.
Record a question about a job you are doing, so Sasha keeps it with the job and the job shows needs_you. Ask the person in Claude Desktop too.
Send your report on a Claude Desktop job back to Sasha when you finish or cannot go on. The job becomes reported, and Sasha starts its own check.
One tool for each skill the organisation enables for external use. A call starts the skill in the background and returns an executionId to poll.
writeDoc and editDoc create and change documents in a member's home or a shared project the person may write to. Off by default on every instance.
surveyPublicContext starts a contained research run that drafts a business brain's Public Context; checkSurvey reports it. Members only, off by default.
One tool for each Sasha App this person can use. A call opens the app's view in a client that shows MCP Apps, and returns the app id and its live revision.
One tool for each action that a Sasha App exposes to the model, such as logging hours. An action that changes data needs a person to confirm it first.
Six tools for each table of a Sasha App that stores data, to list, get, aggregate, create, update and delete rows. A change needs a person to confirm it.
Tools for each report app. Start a run now, read the work log, save the brief, and edit, bin or send a draft (CRM Agent), as the app does.
createProject creates a Sasha workspace for a knowledge project, or confirms it exists, so its documents show in the sidebar. Admin and staff only.
The "overview" section of the Sasha MCP usage guide, as getDocs returns it: What Sasha is and how to work with it.
The "knowledge" section of the Sasha MCP usage guide, as getDocs returns it: The knowledge base.
The "writing" section of the Sasha MCP usage guide, as getDocs returns it: Creating and editing documents.
The "apps-authoring" section of the Sasha MCP usage guide, as getDocs returns it: Editing the source of a Sasha App.
The "permissions" section of the Sasha MCP usage guide, as getDocs returns it: What this connection is allowed to do.
The "skills" section of the Sasha MCP usage guide, as getDocs returns it: Running skills.
The "working-with-bernard" section of the Sasha MCP usage guide, as getDocs returns it: Working alongside Bernard (website editing).
The "limits" section of the Sasha MCP usage guide, as getDocs returns it: Limits and behaviour.
The "improvements" section of the Sasha MCP usage guide, as getDocs returns it: Suggesting improvements, and what happens next.
The "brains" section of the Sasha MCP usage guide, as getDocs returns it: Brains and blueprints.
The "surveys" section of the Sasha MCP usage guide, as getDocs returns it: Research runs for a business brain.
The "reports" section of the Sasha MCP usage guide, as getDocs returns it: Report apps: run now, work log and brief.
The "schedules" section of the Sasha MCP usage guide, as getDocs returns it: Schedules: run a skill at set times.
The "synced-folders" section of the Sasha MCP usage guide, as getDocs returns it: Synced folders: folders copied from GitHub.
The "workspaces" section of the Sasha MCP usage guide, as getDocs returns it: Workspaces: make a knowledge project visible in the sidebar.
The "claude-desktop-jobs" section of the Sasha MCP usage guide, as getDocs returns it: Claude Desktop jobs: clicks in the person's own browser.
The "what-sasha-can-do" section of the Sasha MCP usage guide, as getDocs returns it: What Sasha can do.
The recipe prompts that the Sasha MCP server offers. Each one is a worked example of a common job.
The resource templates of the Sasha MCP server: help articles and the data of Sasha Apps, with the scope each one needs.