You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
- Move the markdown files under /docs/content. This will enable us to
add /docs/blog and other paths in future as needed
- Updated the eleventy link processing logic to handle arbitrary nesting
of the markdown folders.
[TypeAgent Dispatcher](../../ts/packages/dispatcher) is the core component used to build a **personal agent** implementation with _natural language interfaces_. The component's purpose is to understand user requests and translate it into actions using structured prompting and LLM and dispatch them to the appropriate TypeAgent to execute.
8
+
[TypeAgent Dispatcher](../../../ts/packages/dispatcher) is the core component used to build a **personal agent** implementation with _natural language interfaces_. The component's purpose is to understand user requests and translate it into actions using structured prompting and LLM and dispatch them to the appropriate TypeAgent to execute.
9
9
10
10
## Design Goals
11
11
12
12
Here are the design goals of the dispatcher:
13
13
14
14
- Dispatcher processes user requests and use LLM to translate them into actions based on schemas provided by application agents. It automatically finds and switches between different agents to provide a seamless experience in a extensible and scalable way. It aids the user to clarify and complete missing details needed to perform the actions.
15
-
- Dispatcher component can be hosted by different client front ends. [TypeAgent Shell](../../ts/packages/shell) and [TypeAgent CLI](../../ts/packages/cli) are two example of clients built using dispatcher to show case the **personal agent** experience.
16
-
- Dispatcher has extensible [application agents](../../ts/packages/agentSdk/README.md) architecture that allow new agents to be developed and plugin to the **personal agent** experience, scaling up to thousands and more actions.
17
-
- Dispatcher leverage an [agent cache](../../ts/packages/cache/README.md) to lower cost and latency.
18
-
- Dispatcher memorize conversation history by integrating with [KnowPro](../../ts/packages/knowPro) to store past memory and recall for future use.
15
+
- Dispatcher component can be hosted by different client front ends. [TypeAgent Shell](../../../ts/packages/shell) and [TypeAgent CLI](../../../ts/packages/cli) are two example of clients built using dispatcher to show case the **personal agent** experience.
16
+
- Dispatcher has extensible [application agents](../../../ts/packages/agentSdk/README.md) architecture that allow new agents to be developed and plugin to the **personal agent** experience, scaling up to thousands and more actions.
17
+
- Dispatcher leverage an [agent cache](../../../ts/packages/cache/README.md) to lower cost and latency.
18
+
- Dispatcher memorize conversation history by integrating with [KnowPro](../../../ts/packages/knowPro) to store past memory and recall for future use.
19
19
20
-
The current implementation of the [dispatcher](../../ts/packages/dispatcher) is a work in progress toward these goal and have explored most of the pieces described here.
20
+
The current implementation of the [dispatcher](../../../ts/packages/dispatcher) is a work in progress toward these goal and have explored most of the pieces described here.
Copy file name to clipboardExpand all lines: docs/content/architecture/memory.md
+2-2Lines changed: 2 additions & 2 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -25,7 +25,7 @@ Structured RAG is defined as the following steps:
25
25
26
26
Structured RAG can use simple language models to extract entities and topics. This enables Structured RAG to index large conversations, like sets of meeting transcripts. With fine tuning, simple models deliver indices with only a small loss of precision and recall relative to indices built with large language models.
27
27
28
-
The current Structured RAG implementation in the [KnowPro](../../ts/packages/knowPro/README.md) package uses secondary indices for scope expressions such as document range and time range. The implementation also uses secondary indices for related terms, such as "novel" for "book". During query, the memory system discovers related terms and caches them. Models also offer related terms during information extraction.
28
+
The current Structured RAG implementation in the [KnowPro](../../../ts/packages/knowPro/README.md) package uses secondary indices for scope expressions such as document range and time range. The implementation also uses secondary indices for related terms, such as "novel" for "book". During query, the memory system discovers related terms and caches them. Models also offer related terms during information extraction.
29
29
30
30
Structured RAG has the following advantages over state-of-the-art memory using classic RAG:
31
31
@@ -38,7 +38,7 @@ Structured RAG has the following advantages over state-of-the-art memory using c
38
38
39
39
## Implementations
40
40
41
-
- The [KnowPro](../../ts/packages//knowPro/README.md) package (in-development) contains the most recent implementation exploring the ideas of Structured RAG.
41
+
- The [KnowPro](../../../ts/packages//knowPro/README.md) package (in-development) contains the most recent implementation exploring the ideas of Structured RAG.
Copy file name to clipboardExpand all lines: docs/content/help/dev.md
+1-1Lines changed: 1 addition & 1 deletion
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -9,4 +9,4 @@ Here are some tips when working with the TypeAgent repo listed in no particular
9
9
10
10
- When syncing with the [TypeAgent](/) repo it's always a good idea to run `pnpm i` first. This can help resolve some very common build issues or address updated dependency/references within the project hierarchy.
11
11
- If there are issue during `pnpm i` or build, it's a good idea to try resetting the repo to a clean state. Make sure you save all your uncommitted changes (commit them or stash them) and use `git clean` to reset the repo, then try to install and build again.
12
-
- There are two process models for agents: **in-proc** and **out-of-proc** with the [dispatcher](../../ts/packages/dispatcher/). It is recommended to run agents out of process from the dispatcher for system stability possible future isolation. However, this can on occasion make debugging more difficult. Therefore if you are chasing a troublesome issue, try running agents in process the same process as the dispatcher until the issue has been resolved by setting the environment variable `TYPEAGENT_EXECMODE=0`
12
+
- There are two process models for agents: **in-proc** and **out-of-proc** with the [dispatcher](../../../ts/packages/dispatcher/). It is recommended to run agents out of process from the dispatcher for system stability possible future isolation. However, this can on occasion make debugging more difficult. Therefore if you are chasing a troublesome issue, try running agents in process the same process as the dispatcher until the issue has been resolved by setting the environment variable `TYPEAGENT_EXECMODE=0`
0 commit comments