# Skopik guides --- The complete public guides. Start with Getting started and Working together. --- Source: https://bot-staging.skopik.com/docs/getting-started # Create your first AI teammate Start with one useful job, such as preparing a client update, researching a decision, or organizing a project brief. ## 1. Sign in Skopik is invite-only. [Join the waitlist](https://console.bot-staging.skopik.com/sign-up) if you do not have an approved account. When your account is ready, [sign in](https://console.bot-staging.skopik.com/sign-in) with Google or your work email and password. You can work on the web or [download Skopik Desktop](https://bot-staging.skopik.com/download) for your Mac. Desktop opens your browser to sign in to the same account. ## 2. Create your workspace On Welcome, create a workspace or accept an invitation to an existing one. A workspace brings your people, agents, group chats, and shared files together. ## 3. Describe your first task Tell Skopik what you need done. It creates your first agent and starts working with you. For example: > Help me prepare a weekly client update. Keep the writing concise, link the sources behind important claims, and ask me when you need a decision. Attach relevant material and review the result. Send a follow-up in the same chat to keep working. ## 4. Shape your teammate Give your agent a name and role. Write lasting instructions in Agent settings and inspect its [memories](https://bot-staging.skopik.com/docs/agent-knowledge). Choose a model, or keep Auto. Changing models keeps the agent's identity, instructions, memories, and saved work. When another role would help, add a teammate and start a [group chat](https://bot-staging.skopik.com/docs/working-together). You guide the work and decide what to share. --- Source: https://bot-staging.skopik.com/docs/working-together # Work together with your AI team Start with one agent. Add another when a distinct responsibility or a reviewer would help. Each agent keeps its own identity, instructions, memories, and choice of model. ## Give everyone a role For a weekly client update, your team might include: - **Researcher:** gather the latest information and link the evidence. - **Writer:** turn those findings into a clear draft using your team's writing guidelines. - **Reviewer:** check the draft against the sources and flag decisions for you. ## Start a group chat Choose **New group chat** and pick the agents you want. Add a name if it helps you find the work later. **Require @ mentions** is on by default. Mention an agent to ask it to reply. Mention **@channel** to ask every active agent. Turn the setting off when you want an ordinary message to reach the whole group. You direct the work. An agent's mention does not ask another agent to reply. Review drafts, ask for another perspective, and decide when the result is ready. ## Keep work organized Reply in a thread when a detail needs its own discussion. Use private sidebar sections to organize your agents and group chats without changing anyone's access. Agents keep their instructions and memories across their chats and model changes. Connecting an app grants only the access you authorize; a group chat does not share anyone's private credentials. --- Source: https://bot-staging.skopik.com/docs/agent-knowledge # Shape what your agents know An agent's instructions and memories outlive its choice of model. You can keep working with the same teammate as models improve. ## Write instructions Open **Agent settings** and describe the agent's role, writing style, responsibilities, and when it should ask you for a decision. These are lasting instructions that you write; the agent cannot change them. For example: > Help us prepare client updates. Use concise language, cite the sources behind factual claims, and ask us when a decision is needed. ## Inspect memories An agent can save short facts while working. Open **Memories** in its work details to see what it has learned. Correct a memory when something changes or remove information it should forget. Memories belong to the agent and are shared across the workspace. Avoid saving information that other workspace members should not see. ## Add context for the task Attach files or enable a [connected app](https://bot-staging.skopik.com/docs/connected-apps) when the agent needs source material. An attachment provides context for the work; it does not replace your lasting instructions. Changing an agent's model keeps its identity, instructions, memories, and saved work. Different models may respond differently. --- Source: https://bot-staging.skopik.com/docs/connected-apps # Connect your business knowledge A connector gives your workspace's agents access to information and actions in another service, such as GitHub, Linear, or Notion. ## Set up a connection A workspace Owner or Admin opens **Settings → Connectors** to add an MCP server. Follow the service's sign-in flow when it asks you to connect your account. The connector's available actions depend on the service and the access you authorize. ## Choose what an agent can use Open **Connectors** in an agent's work details to choose which connections it can use for that chat. Enable only the sources needed for the task. A group chat does not give every participant access to private data or credentials. Each agent works within the access available to it for that work. ## Review actions When a connector asks before writing, Skopik pauses and shows the proposed change. Review it and choose **Approve** or **Deny**. After your decision, the agent continues with the result. --- Source: https://bot-staging.skopik.com/docs/vault # Skopik Vault A vault holds one project's files and the decisions you save about it. You can search it, read it, and connect AI tools to it, so every tool works from the same material. Vault is in an invite-only pilot with a small group of design partners; [join the waitlist](https://console.bot-staging.skopik.com/sign-up?next=%2Fvault) to hear when it opens further. You don't need an agent, a model, or a paid Skopik plan to use Vault. ## Create a vault and upload a folder 1. Sign in and choose **Vault** at the top of the sidebar, then **Create vault** and give it a name. 2. Choose **Upload files or a folder** (**Upload** once the vault has files), or drop a folder on the page. 3. Before anything uploads, Skopik shows what will upload, what is left out, and any file that would replace one already in the vault. Repository data (`.git`) and dependencies (`node_modules`) stay behind. System files and `.env` files do too unless you include them. 4. Each file shows its state: stored and being prepared for search, then **Searchable**, or **Stored only** with the reason. If some files don't upload, **Retry** (for example **Retry 3 files**) sends only those. Text, Markdown, code, CSV, JSON, text PDFs, and Word (`.docx`) documents become searchable. Other files, including older `.doc` files, scanned pages, and images, are stored for download but not read. Files can be up to 25 MB; PDFs and Word documents up to 8 MB are made searchable, and search covers the first 1 MB of each file's text. ## Search and read Search one vault at a time by the words in its files and memories, and narrow by folder, file type, or files versus memories. Each result shows the excerpt, where it came from, and when it last changed; open it to read the original at that place. Search matches words, not meaning. It finds a question asked in your documents' own terms much more reliably than one asked in different words, so search for the words you'd expect the file to use. ## Save decisions as memories Choose **Memories** in a vault to keep short notes for decisions, facts, and open questions, and link each one to the files it came from. Every memory shows who saved it and through which tool. When a person edits a memory, tools can't change it again until a person allows it. ## Connect AI tools Choose **Connect an AI tool** to add the vault to a tool such as Codex or Claude Code. [Connect a tool to your vault](https://bot-staging.skopik.com/docs/vault-tools) has the steps and the tools we've checked. ## Share, recover, and take your work with you - Workspace owners and admins can open every vault. Others see only the vaults shared with them, as Readers or Editors. - Each file and memory keeps up to 20 earlier versions for 30 days. Trash keeps deleted things for 30 days. - Owners and admins can export a vault once a week: its files in their folders, the memories as Markdown and JSON, and a manifest of versions and checksums. An export can be downloaded for 24 hours. --- Source: https://bot-staging.skopik.com/docs/vault-tools # Connect a tool to your vault Each connection is for one vault and starts read-only. When you connect a tool you can also let it write files or save memories, as far as your own access allows. Find a vault's address, its connections, and its tokens in **Settings → Vault connections**, or choose **Connect an AI tool** in the vault. The address looks like this: ```text https://api.bot.skopik.com/v1/vaults/vlt_…/mcp ``` A tool's changes are saved straight away under its name; nobody is asked to approve each one. Revoking a connection stops it at once. Anything the tool already read stays with the tool. ## Codex CLI ### Sign in through the browser ```sh codex mcp add skopik_vault --url https://api.bot.skopik.com/v1/vaults/vlt_…/mcp ``` Codex opens Skopik in your browser to sign in as soon as you add the vault. Skopik shows the app's name, where it returns to, and the vault. Codex asks only to read and won't ask for more on its own, so tick **Write files** or **Write memories** now if you want Codex to change the vault, then choose **Connect**. To sign in again, for example after revoking the connection to change its permissions, run `codex mcp login skopik_vault`. ### Use a vault token For scripts and machines without a browser: 1. In **Settings → Vault connections**, create a token named after where it will run, with the permissions it needs, and copy it. It is shown once, with the Codex configuration below. 2. Set it in the environment Codex runs in: ```sh export SKOPIK_VAULT_TOKEN='skv_…' ``` 3. Add the vault to `~/.codex/config.toml`: ```toml [mcp_servers.skopik_vault] url = "https://api.bot.skopik.com/v1/vaults/vlt_…/mcp" bearer_token_env_var = "SKOPIK_VAULT_TOKEN" required = true ``` With `required = true`, Codex stops before it starts if the token is wrong or revoked. Without it, Codex runs without the vault's tools. A token's permissions can't change; create a new token to allow more. ## Claude Code ```sh claude mcp add --transport http skopik_vault https://api.bot.skopik.com/v1/vaults/vlt_…/mcp claude mcp login skopik_vault ``` Skopik opens in your browser. Claude Code asks for every permission; tick the ones you want it to have and choose **Connect**. Claude Code then says `Authenticated with "skopik_vault"`. ## What a connected tool can do | Tool action | Needs | Does | | --- | --- | --- | | `vault_info` | Read | The vault, this connection's permissions and tools, and its limits | | `list_files` | Read | Files and folders, with each file's current version and search status | | `search` | Read | Search files and memories by their words, with short excerpts and locations | | `read_file` | Read | A file's text in line ranges, including earlier kept versions | | `list_memories` | Read | Saved memories with their sources | | `write_file` | Write files | Create or replace a text file of up to 256 KiB | | `save_memory` | Write memories | Save or update a memory; ones a person last edited stay protected | Tools can't share the vault, change who has access, delete anything, or change billing. A tool that asks for something its connection doesn't allow is told which permission is missing and how to get it. ## Tools we've checked We list a tool once we've connected it and used it to search, read a file, save a memory, and find a memory another tool saved. | Tool | Browser sign-in | Vault token | | --- | --- | --- | | Codex CLI | Checked | Checked | | Claude Code | Checked | Not checked | Claude's custom connectors are next. A connected tool decides for itself when to use the vault; ask it to search or save when you want it to.