Platform for stateful agents: AI with advanced memory that can learn and self-improve over time.
bump version
Caren Thomas72871ff9231a
diff --git a/fern/pages/cloud/templates.mdx b/fern/pages/cloud/templates.mdxdeleted file mode 100644index 179342765..000000000--- a/fern/pages/cloud/templates.mdx+++ /dev/null@@ -1,131 +0,0 @@-----title: Introduction to Agent Templates-slug: guides/templates/overview------<Tip>-Agent Templates are a feature in [Letta Cloud](/guides/cloud) that allow you to quickly spawn new agents from a common agent design.-</Tip>--Agent templates allow you to create a common starting point (or *template*) for your agents.-You can define the structure of your agent (its tools and memory) in a template,-then easily create new agents off of that template.--<Frame>-```mermaid-flowchart TD- subgraph Template["Agent Template v1.0"]- tools["Custom Tools- --------- tool_1- tool_2- tool_3"]- memory["Memory Structure- ---------------- system_instructions- core_memory- archival_memory"]- end-- Template --> |Deploy| agent1["Agent 1- --------- Custom state"]- Template --> |Deploy| agent2["Agent 2- --------- Custom state"]- Template --> |Deploy| agent3["Agent 3- --------- Custom state"]-- class Template template- class agent1,agent2,agent3 agent-```-</Frame>--Agent templates support [versioning](/guides/templates/versioning), which allows you to programatically-upgrade all agents on an old version of a template to the new version of the same template.--Agent templates also support [memory variables](/guides/templates/variables), a way to conveniently customize-sections of memory at time of agent creation (when the template is used to create a new agent).--## Agents vs Agent Templates--**Templates** define a common starting point for your **agents**, but they are not agents themselves.-When you are editing a template in the ADE, the ADE will simulate an agent for you-(to help you debug and design your template), but the simulated agent in the simulator is not retained.--You can refresh the simulator and create a new simulated agent from your template at any time by clicking the "Flush Simulation" button 🔄 (at the top of the chat window).--To create a persistent agent from an existing template, you can use the [create agents from template endpoint](/api-reference/templates/agents/create):-```sh-curl -X POST https://app.letta.com/v1/templates/{template_name}:{template_version} \- -H 'Content-Type: application/json' \- -H 'Authorization: Bearer YOUR_API_KEY' \- -d '{}'-```--### Creating a template from an agent-You may have started with an agent and later decide that you'd like to convert it into a template to allow you to easily create new copies of your agent.--To convert an agent (deployed on Letta Cloud) into a template, simply open the agent in the ADE and click the "Convert to Template" button.--## Example usecase: customer service-Imagine you're creating a customer service chatbot application.-You may want every user that starts a chat sesion to get their own personalized agent:-the agent should know things specific to each user, like their purchase history, membership status, and so on.--<Frame>-```mermaid-flowchart TD- subgraph Template["Customer Service Template"]- tools["Custom Tools- --------- update_ticket_status- search_knowledge_base- escalate_ticket"]- memory["Memory Structure- ---------------- name: {{name}}- ticket: {{ticket}}- spent: {{amount}}"]- end-- Template --> |Deploy| user1["Alice's Agent- --------- name: Alice- ticket: T123- spent: $500"]- Template --> |Deploy| user2["Bob's Agent- --------- name: Bob- ticket: T124- spent: $750"]- Template --> |Deploy| user3["Carol's Agent- --------- name: Carol- ticket: T125- spent: $1000"]-- class Template template- class user1,user2,user3 agent-```-</Frame>--However, despite being custom to individual users, each agent may share a common structure:-all agents may have access to the same tools, and the general strucutre of their memory may look the same.-For example, all customer service agents may have the `update_ticket_status` tool that allows the agent to update the status of a support ticket in your backend service.-Additionally, the agents may share a common structure to their memory block storing user information.--This is the perfect scenario to use an **agent template**!--You can take advantage of memory variables to write our user memory (one of our core memory blocks) to exploit the common structure across all users:-```handlebars-The user is contacting me to resolve a customer support issue.-Their name is {{name}} and the ticket number for this request is {{ticket}}.-They have spent ${{amount}} on the platform.-If they have spent over $700, they are a gold customer.-Gold customers get free returns and priority shipping.-```--Notice how the memory block uses variables (wrapped in `{{ }}`) to specify what part of the memory should be defined at agent creation time, vs within the template itself.-When we create an agent using this template, we can specify the values to use in place of the variables.diff --git a/fern/pages/cloud/variables.mdx b/fern/pages/cloud/variables.mdxdeleted file mode 100644index 6d5f5c597..000000000--- a/fern/pages/cloud/variables.mdx+++ /dev/null@@ -1,54 +0,0 @@-----title: Memory Variables-slug: guides/templates/variables------<Note>-Memory variables are a feature in [agent templates](/guides/cloud/templates) (part of [Letta Cloud](/guides/cloud)).-To use memory variables, you must be using an agent template, not an agent.-</Note>--Memory variables allow you to dynamically define parts of your agent memory at the time of agent creation (when a [template](/guides/cloud/templates) is used to create a new agent).--## Defining variables in memory blocks--To use memory variables in your agent templates, you can define variables in your memory blocks by wrapping them in `{{ }}`.-For example, if you have an agent template called `customer-service-template` designed to handle customer support issues, you might have a block of memory that stores information about the user:-```handlebars-The user is contacting me to resolve a customer support issue.-Their name is {{name}} and the ticket number for this request is {{ticket}}.-```--Once variables have been defined inside of your memory block, they will dynamically appear at variables in the **ADE variables window** (click the "\{\} Variables" button at the top of the chat window to expand the dropdown).--## Simulating variable values in the ADE--<Tip>-Reset the state of the simulated agent by clicking the "Flush Simulation" 🔄 button.-</Tip>--While designing agent templates in the ADE, you can interact with a simulated agent.-The ADE variables window allows you to specify the values of the variables for the simulated agent.--You can see the current state of the simulated agent's memory by clicking the "Simulated" tab in the "Core Memory" panel in the ADE.-If you're using memory variables and do not specify values for the variables in the ADE variables window, the simulated agent will use empty values.--In this prior example, the `name` and `ticket` variables are memory variables that we will specify when we create a new agent - information that we expect to have available at that time.-While designing the agent template, we will likely want to experiment with different values for these variables to make sure that the agent is behaving as expected.-For example, if we change the name of the user from "Alice" to "Bob", the simulated agent should respond accordingly.--## Defining variables during agent creation--When we're ready to create an agent from our template, we can specify the values for the variables using the `variables` parameter in the [create agents from template endpoint](/api-reference/templates/agents/create):-```sh-curl -X POST https://app.letta.com/v1/templates/{template_name}:{template_version} \- -H 'Content-Type: application/json' \- -H 'Authorization: Bearer YOUR_API_KEY' \- -d '{- "from_template": customer-service-template:latest",- "variables": {- "name": "Bob",- "ticket": "TX-123"- }- }'-```diff --git a/fern/pages/cloud/versions.mdx b/fern/pages/cloud/versions.mdxdeleted file mode 100644index 905e64c60..000000000--- a/fern/pages/cloud/versions.mdx+++ /dev/null@@ -1,41 +0,0 @@-----title: Versioning Agent Templates-slug: guides/templates/versioning------<Note>-Versioning is a feature in [agent templates](/guides/cloud/templates) (part of [Letta Cloud](/guides/cloud/overview)).-To use versioning, you must be using an agent template, not an agent.-</Note>--Versions allow you to keep track of the changes you've made to your template over time.-Agent templates follow the versioning convention of `template-name:version-number`.--Similar to [Docker tags](https://docs.docker.com/get-started/docker-concepts/building-images/build-tag-and-publish-an-image/#tagging-images), you can specify the latest version of a template using the `latest` keyword (`template-name:latest`).--## Creating a new template version-When you create a template, it starts off at version 1.-Once you've make edits to your template in the ADE, you can create a new version of the template by clicking the "Template" button in the ADE (top right), then clicking "Save new template version".-Version numbers are incremented automatically (e.g. version 1 becomes version 2).--## Migrating existing agents to a new template version-If you've deployed agents on a previous version of the template, you'll be asked if you want to migrate your existing agents to the new version of the template.-When you migrate existing agents to a new template version, Letta Cloud will re-create your existing agents using the new template information, but keeping prior agent state such as the conversation history, and injecting memory variables as needed.--### When should I migrate (or not migrate) my agents?-One reason you might want to migrate your agents is if you've added new tools to your agent template: migrating existing agents to the new version of the template will give them access to the new tools, while retaining all of their prior state.-Another example usecase is if you make modifications to your prompts to tune your agent behavior - if you find a modification works well, you can save a new version with the prompt edits, and migrate all deployed agents to the new version.--### Forking an agent template-If you decide to make significant changes to your agent and would prefer to make a new template to track your changes, you can easily create a new agent template from an existing template by **forking** your template (click the settings button ⚙️ in the ADE, then click "Fork Template").--## Specifying a version when creating an agent--You can specify a template version when creating an agent in the you can use the [create agents from template endpoint](/api-reference/templates/agents/create)-For example, to deploy an agent from a template called `template-name` at version 2, you would use `:2` as the template tag:-```sh-curl -X POST https://app.letta.com/v1/templates/{template_name}:2 \- -H 'Content-Type: application/json' \- -H 'Authorization: Bearer YOUR_API_KEY' \- -d '{}'-```diff --git a/letta/__init__.py b/letta/__init__.pyindex 92224bcda..5019bbad9 100644--- a/letta/__init__.py+++ b/letta/__init__.py@@ -5,7 +5,7 @@ try:__version__ = version("letta")except PackageNotFoundError:# Fallback for development installations- __version__ = "0.16.3"+ __version__ = "0.16.4"if os.environ.get("LETTA_VERSION"):__version__ = os.environ["LETTA_VERSION"]diff --git a/pyproject.toml b/pyproject.tomlindex 9d5722df3..a61193836 100644--- a/pyproject.toml+++ b/pyproject.toml@@ -1,6 +1,6 @@[project]name = "letta"-version = "0.16.3"+version = "0.16.4"description = "Create LLM agents with long-term memory and custom tools"authors = [{name = "Letta Team", email = "[email protected]"},diff --git a/uv.lock b/uv.lockindex a16099adb..8028a7a91 100644--- a/uv.lock+++ b/uv.lock@@ -2510,7 +2510,7 @@ wheels = [[[package]]name = "letta"-version = "0.16.2"+version = "0.16.4"source = { editable = "." }dependencies = [{ name = "aiofiles" },