> ## Documentation Index
> Fetch the complete documentation index at: https://docs.karta.sh/llms.txt
> Use this file to discover all available pages before exploring further.
# Choose your path
> Find the right route through the Karta docs for your current job: first deploy, existing harness project, embedded widget, backend API, operations, or enterprise review.
Karta has a small core model, but readers arrive with different jobs. Start with
the path that matches the thing you need to ship, integrate, operate, or review.
If you already know the task in your own words, use the
[How-to index](/how-to). It lists the public process questions directly, so
docs search and agent retrieval can find the canonical answer.
## Fast paths
| If you need to... | Use this path | Primary surface | Success signal |
| --------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------- | ----------------------------------------------------------------------------------- |
| Deploy your first agent project | [Quickstart](/quickstart), then [Deploy loop](/deploy/deploy-loop) | `karta create`, `karta setup --enable`, `karta deploy`, `karta open` | `karta open` shows the hosted chat page for a live release. |
| Bring an existing supported-harness project | [Supported harnesses](/build/supported-harnesses), then [Agent structure](/build/agent-structure) and [karta.toml](/build/karta-toml) | Existing harness folder plus `karta.toml` | The folder has a deploy manifest and ships as a release. |
| Build a local integration before cloud deploy | [Support-bot tutorial](/tutorials/support-bot) | `karta dev`, local session API, widget | A browser widget streams from the local Karta surface. |
| Add a public chat widget to a site | [Chat widget quickstart](/sdks/widget/quickstart) | Publishable embed key, allowed origins, script tag | A launcher appears on your site and streams replies. |
| Serve signed-in users with persistent memory | [Identity](/sdks/widget/identity), then [Authenticated agent](/sdks/widget/authenticated-agent) | Verified identity, session token, stable karta id | Each verified user resumes the correct durable karta. |
| Drive an agent from your backend | [API reference](/api-reference/introduction) | API key, `/sessions`, `/messages`, SSE | Your server creates a session and sends a streamed turn. |
| Reuse an OpenAI or Anthropic-shaped client | [Consumer adapters](/api-reference/adapters/overview) | `/responses`, `/chat/completions`, managed-agent event log | Existing client calls reach a Karta agent. |
| Choose the agent-instance model | [Kartas & memory](/concepts/instances-and-memory) | `agent_instance_id`, `end_user_id`, session metadata | You can name the id for each user, virtual employee, backend task, or fleet member. |
| Review production architecture and security | [Production architecture review](/security/architecture-review) | Runtime/management boundary, credentials, spend controls, audit | You can explain the trust boundary and launch controls. |
| Operate agents in production | [Production readiness](/platform/production-readiness) | Releases, logs, budgets, webhooks, billing, audit | You can deploy, roll back, inspect sessions, cap spend, and receive events. |
## By role
Install the CLI, scaffold an agent project, deploy it, and open the live chat page.
Keep the supported harness project format you already use and add Karta's
deploy layer.
Drop in the hosted widget, customize it, and wire commands or identity later.
Pick the native session API or an adapter shape your current client already speaks.
Review isolation, credentials, spend gates, audit, webhooks, and production controls.
Manage keys, releases, budgets, billing, webhooks, audit, and troubleshooting.
## Integration decisions
Make these choices before you write integration code.
### 1. How will users reach the agent?
* Use the [hosted chat widget](/sdks/widget/quickstart) when you want the
fastest customer-facing UI.
* Use the [native session API](/api-reference/sessions) when your backend owns
the product experience.
* Use a [consumer adapter](/api-reference/adapters/overview) when your code is
already written around an OpenAI or Anthropic-shaped client.
### 2. Who is allowed to hold the credential?
* Server-side code can hold a scoped [`kt_live_` API key](/platform/api-keys).
* Browsers should receive only a short-lived [session token](/security/session-tokens)
or a publishable widget key fenced by allowed origins.
### 3. What should persist?
The stable karta id selects the agent instance, workspace, and memory. Pick one
karta per user, one per virtual employee, one per backend task, or one per fleet
member. See [Kartas & memory](/concepts/instances-and-memory).
### 4. How will you operate it?
Before launch, decide who can deploy, how rollback works, what budget caps apply,
which events should reach your systems, and what your support team should inspect
when a session fails. Start with [Production readiness](/platform/production-readiness)
and [Production architecture review](/security/architecture-review).