itermbot 1.0.15 → 1.0.16
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/config/memory.yaml +51 -16
- package/package.json +1 -1
package/config/memory.yaml
CHANGED
|
@@ -3,37 +3,72 @@ kind: MemoryConfig
|
|
|
3
3
|
metadata:
|
|
4
4
|
name: itermbot-memory
|
|
5
5
|
spec:
|
|
6
|
-
|
|
6
|
+
types:
|
|
7
7
|
thread:
|
|
8
8
|
store:
|
|
9
9
|
type: in_memory
|
|
10
|
-
|
|
11
|
-
|
|
10
|
+
# Empty model means using model.yaml defaults.
|
|
11
|
+
# You can override per type, e.g. llmModelId/embedModelId.
|
|
12
|
+
model: {}
|
|
13
|
+
keywords:
|
|
14
|
+
- thread
|
|
15
|
+
- session
|
|
16
|
+
- conversation
|
|
12
17
|
sessionCompaction:
|
|
13
18
|
enabled: true
|
|
14
19
|
maxTokens: 3000
|
|
15
20
|
keepRecentTurns: 12
|
|
16
21
|
summaryKey: __session_summary__
|
|
17
22
|
checkEveryTurns: 3
|
|
23
|
+
transformContent:
|
|
24
|
+
enabled: false
|
|
18
25
|
|
|
19
26
|
cross_thread:
|
|
20
27
|
store:
|
|
21
28
|
type: in_memory
|
|
22
|
-
|
|
23
|
-
|
|
29
|
+
# Empty model means using model.yaml defaults.
|
|
30
|
+
# You can override per type, e.g. llmModelId/embedModelId.
|
|
31
|
+
model: {}
|
|
32
|
+
keywords:
|
|
33
|
+
- preference
|
|
34
|
+
- profile
|
|
35
|
+
- user
|
|
36
|
+
- setting
|
|
37
|
+
transformContent:
|
|
38
|
+
enabled: false
|
|
39
|
+
systemPrompt: |
|
|
40
|
+
Rewrite into one durable user preference/fact sentence.
|
|
41
|
+
Keep only stable user attributes, constraints, and likes/dislikes.
|
|
42
|
+
Output only the rewritten memory text.
|
|
24
43
|
|
|
25
44
|
knowledge:
|
|
26
45
|
store:
|
|
27
46
|
type: in_memory
|
|
28
|
-
|
|
29
|
-
|
|
47
|
+
# Empty model means using model.yaml defaults.
|
|
48
|
+
# You can override per type, e.g. llmModelId/embedModelId.
|
|
49
|
+
model: {}
|
|
50
|
+
keywords:
|
|
51
|
+
- /knowledge/
|
|
52
|
+
- /docs/
|
|
53
|
+
- knowledge
|
|
54
|
+
- .md
|
|
55
|
+
- .pdf
|
|
56
|
+
- api
|
|
57
|
+
- reference
|
|
58
|
+
transformContent:
|
|
59
|
+
enabled: false
|
|
60
|
+
systemPrompt: |
|
|
61
|
+
Rewrite into one concise factual knowledge sentence.
|
|
62
|
+
Preserve entity names, APIs, versions, and constraints.
|
|
63
|
+
Output only the rewritten memory text.
|
|
30
64
|
|
|
31
|
-
|
|
32
|
-
|
|
33
|
-
|
|
34
|
-
|
|
35
|
-
|
|
36
|
-
|
|
37
|
-
-
|
|
38
|
-
-
|
|
39
|
-
-
|
|
65
|
+
classification:
|
|
66
|
+
# Empty model means using model.yaml llm default.
|
|
67
|
+
# You can override classifier model via llmModelId.
|
|
68
|
+
model: {}
|
|
69
|
+
systemPrompt: |
|
|
70
|
+
Classify the memory into exactly one label:
|
|
71
|
+
- thread: session-local temporary context.
|
|
72
|
+
- cross_thread: durable user preference/fact across sessions.
|
|
73
|
+
- knowledge: stable reference/doc/API knowledge.
|
|
74
|
+
Return exactly one token: thread | cross_thread | knowledge.
|