> For the complete documentation index, see [llms.txt](https://notara.gitbook.io/notara-docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://notara.gitbook.io/notara-docs/faq.md). # Frequently Asked Questions ## Setup & Account ### Why is BYOK required? Can't I just pay for access to a model? Notara is a platform, not a model. We connect your team's context and tools to an AI model — but we deliberately don't bundle the model. This gives you: * **Cost transparency**: your provider bills you directly. There's no opaque "AI usage" line item in your Notara invoice. * **Provider choice**: you pick the model that works best for your team and compliance requirements. * **Your rate limits**: you're not sharing capacity with other Notara customers. * **No markup**: we don't take a cut of your token spend. Anthropic Claude is the recommended provider for best performance, but Notara works with OpenAI, Gemini, Kimi, GLM, and DeepSeek as well. ### Can I use Notara without Slack? The web dashboard and MCP endpoint work without Slack. You can manage libraries, skills, connectors, and automations entirely from the dashboard. The MCP endpoint lets external agents (Claude Code, Cursor, etc.) access your workspace. The Slack bot is how most teams get day-to-day value from Notara — it's the primary conversation interface — but it's not required to use other features. ### I signed up with the wrong Slack workspace. How do I fix it? Go to **Settings → Workspace → Slack Integration**, click **Disconnect**, then **Connect Slack Workspace** and authorize the correct workspace. Your data (libraries, skills, automations) is tied to your Notara workspace, not to the Slack workspace, so nothing is lost. ### Can I connect multiple Slack workspaces to one Notara workspace? Not currently. Each Notara workspace connects to one Slack workspace. If you manage multiple Slack workspaces, you'll need separate Notara workspaces for each. You can belong to multiple Notara workspaces with a single login. *** ## Credits & Billing ### How does credit counting work exactly? Credits = (input tokens + output tokens) / 10,000. Only calls to your primary agent model count — triage/classification calls to lightweight routing models are excluded. 1 credit = 10,000 tokens = roughly 7,500 words. ### My credits ran out faster than expected. What happened? Check **Settings → Billing → Usage** for a day-by-day breakdown. Common causes: * Long spine documents injected on every turn (switch large docs to `on_demand`) * Automations running frequently with large contexts * Long Slack conversation threads accumulating history ### Do unused credits roll over? No. Credits reset on your billing cycle date. This is standard for most SaaS credit systems — the included credits represent a monthly capacity, not an accumulating balance. ### Is my LLM usage private from Notara? Yes. Your BYOK key calls your provider's API directly from Notara's servers. Notara logs the **token counts** for credit accounting, but never logs the content of prompts or responses. Your conversation content is not accessible to Notara staff. *** ## Tools & Permissions ### How does the approval flow work in practice? When a team member asks the agent to do something with a `write` tool (like creating a Linear issue), the agent pauses and sends a Block Kit message to the first admin's Slack DM. The admin sees the proposed action and clicks Approve or Deny. Approving executes the tool; denying cancels it with the reason posted to the original thread. The flow expires after 30 minutes if the admin doesn't respond. ### Can I turn off the approval flow for a specific tool? Yes. Go to **Settings → Permissions**, find the tool, and change its permission level from `requires_approval` to `open`. This makes the tool execute immediately without approval. Use this for tools where your team has high trust and the action is easily reversible. You can also make this change only for specific channels using per-channel permission overrides, while keeping the workspace default at `requires_approval`. ### What's the difference between `write` and `destructive` tools? `write` tools create or modify data (create a Linear issue, send a message, update a record). `destructive` tools delete or irreversibly change data (void an invoice, delete a file, mass-update records). Destructive tools require admin-only access AND an in-channel confirmation prompt for every execution — even for admins. This extra layer prevents prompt injection attacks where someone tricks the agent into a destructive action. *** ## Skills & Libraries ### What's the difference between a skill and an automation? A **skill** is a prompt configuration — it tells the agent *how* to do something when invoked. An **automation** is a *trigger* — it schedules when something runs or connects an external event to a run. You often use both together: an automation triggers daily, and it invokes a skill that defines how to do the weekly digest. ### Can team members with the editor role manage skills without an admin? Yes. Editors can create, edit, and fork skills. The only limit is that editors cannot grant a skill elevated tool permissions — if a workspace has `linear_create_issue` at `requires_approval`, an editor can't change the skill to bypass that. Granting elevated access requires admin. ### How many skill versions are kept? Unlimited. Every save creates a version. You can browse and restore any previous version from the skill's version history panel. ### What's the `never` load policy for? Can the agent still find those documents? No — `never` means the agent cannot see the document at all. It's not in the manifest, not indexed for semantic search, and not accessible via MCP. It's purely for human reference: internal notes, sensitive drafts, historical records you don't want the agent to retrieve or reference. ### Is there a limit on how many documents a library can hold? No hard limit. Large libraries (hundreds of documents) work best with `on_demand` load policies — spine documents scale poorly because they're injected on every turn. With `on_demand`, the agent retrieves only what's relevant, so library size doesn't meaningfully affect performance. *** ## MCP & External Agents ### What can an external agent do via MCP that it can't do without it? Via MCP, an external agent (Claude Code, Cursor, a custom agent) gets access to your workspace's full tool set (connector tools), context library (as MCP resources), and skills (as MCP prompts). Without Notara MCP, the agent only has its own training data and whatever tools the IDE provides natively. ### Are MCP callers as trusted as Slack users? MCP callers are never treated as system actors or admins. They're subject to the same permission system as regular team members. Write tools require approval, destructive tools require admin confirmation. The permission level of an MCP caller is determined by the group configuration for the endpoint — it cannot self-elevate. *** ## Support ### How do I get help? Email with a description of your issue and your workspace name. Response time is typically within 1 business day. For urgent issues affecting production, include "URGENT" in the subject line. ### Is there a status page? Yes — check [status.notara.ai](https://status.notara.ai) for real-time service status and incident history. --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter: ``` GET https://notara.gitbook.io/notara-docs/faq.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.