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
{{ message }}
Repository navigation
Commit f8ea416
Browse filesBrowse the repository at this point in the historyBrowse files
Copy file name to clipboardExpand all lines: docs/guide/debug-ui.md
+34-51Lines changed: 34 additions & 51 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -1,6 +1,9 @@
1
1
# Chat Interface (Chainlit)
2
2
3
-
Agentomatic includes a built-in ChatGPT-like conversational interface powered by **Chainlit**. It provides an interactive chat playground to test agent responses, compare prompt versions, inspect tool calls, and collect user feedback — without writing any frontend code.
3
+
Agentomatic includes a built-in Chainlit **debug chat**. It lets an operator
4
+
select a deployed agent, send a synchronous invoke request, and inspect the
5
+
returned response, execution-step labels, citations, and suggestions — without
6
+
writing a frontend.
4
7
5
8
!!! info "Chat UI vs Agentomatic Studio"
6
9
Agentomatic offers **two** debug interfaces for different workflows:
@@ -10,7 +13,7 @@ Agentomatic includes a built-in ChatGPT-like conversational interface powered by
Mounts the Chainlit interface directly into your FastAPI application. API requests and chat sessions share the same persistence backend and middleware stack.
36
+
Mounts the Chainlit interface into the FastAPI application. The debug chat
37
+
invokes the platform's normal `/invoke` API, so that API request follows
38
+
the platform middleware.
34
39
35
40
```bash
36
41
agentomatic run --with-ui
@@ -71,69 +76,45 @@ pip install "agentomatic[ui]"
71
76
agentomatic run --with-ui --reload
72
77
```
73
78
79
+
!!! warning "Local debug surface, not an authenticated operator console"
80
+
The bundled UI currently sends plain `/invoke` requests and does not
81
+
forward an API key, JWT, browser identity, or a conversation `thread_id`.
82
+
It is intended for a local or otherwise trusted debugging setup. For a
83
+
production frontend, use the documented REST/Studio APIs and your own
84
+
authentication and session handling.
85
+
74
86
---
75
87
76
88
## Interface Features
77
89
78
90
### :material-robot: Agent Selector
79
91
80
-
A top-navigation dropdown lists all registered agents discovered by the platform registry. Select an agent to dynamically load its input form, configuration, and documentation.
81
-
82
-
### :material-tag: Prompt Version Selector
83
-
84
-
Inspect and switch between prompt versions (e.g., `v1`, `v2`, `v1_formal`) on the fly. Chat queries execute against the selected version, enabling manual A/B comparison of prompt behaviors.
85
-
86
-
```json
87
-
// agents/my_agent/prompts.json
88
-
{
89
-
"v1": {
90
-
"system": "You are a concise assistant.",
91
-
"user_template": "Query: {query}"
92
-
},
93
-
"v2": {
94
-
"system": "You are a creative, detailed assistant.",
95
-
"user_template": "Please elaborate on: {query}"
96
-
}
97
-
}
98
-
```
99
-
100
-
### :material-waves: Token-by-Token Streaming
101
-
102
-
If your agent supports streaming (via SSE), response completions stream onto the screen in real-time, matching the experience of ChatGPT and similar interfaces.
A Chainlit settings selector lists agents returned by the platform registry.
93
+
Select an agent to send subsequent messages to that agent's `/invoke` route.
105
94
106
-
Intermediate agent actions — tool calls, function invocations, retrieval steps — are captured and rendered as clean, expandable panels in the chat flow. Click a panel to inspect the exact input arguments and JSON output returned by the tool.
If your agent returns reasoning or step-by-step logs, the UI highlights these in collapsible cards showing the agent's thought process before the final answer.
97
+
Each message creates an expandable Chainlit step containing the `/invoke`
98
+
payload and complete JSON response. If the response contains `steps_taken`,
99
+
the UI displays those agent-supplied execution labels in a second step. It does
100
+
not stream `/invoke/stream`, inspect provider tool calls independently, or
Citations returned by the agent are displayed as an expandable JSON retrieval
106
+
step. The bundled UI does not submit feedback; use the agent feedback API or a
107
+
production client for that workflow.
129
108
130
109
---
131
110
132
111
## Customization
133
112
134
113
### Theme & Layout
135
114
136
-
When running `agentomatic run --with-ui` for the first time, Agentomatic generates a default `.chainlit/config.toml` file. Customize it to match your brand:
115
+
Chainlit controls its own theme and layout files. If your project has a
116
+
`.chainlit/config.toml`, customize it according to your installed Chainlit
117
+
version; Agentomatic does not generate or manage that file.
0 commit comments