entroly-openclaw 1.0.78 → 1.0.80
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/README.md +24 -4
- package/package.json +1 -1
package/README.md
CHANGED
|
@@ -22,7 +22,7 @@ Install the Python engine, then install the publicly verifiable OpenClaw plugin
|
|
|
22
22
|
package from npm:
|
|
23
23
|
|
|
24
24
|
```bash
|
|
25
|
-
pip install "entroly>=1.0.
|
|
25
|
+
pip install "entroly>=1.0.80"
|
|
26
26
|
openclaw plugins install npm:entroly-openclaw
|
|
27
27
|
openclaw plugins enable entroly
|
|
28
28
|
openclaw gateway restart
|
|
@@ -36,7 +36,7 @@ installable release.
|
|
|
36
36
|
From an Entroly source checkout:
|
|
37
37
|
|
|
38
38
|
```bash
|
|
39
|
-
pip install "entroly>=1.0.
|
|
39
|
+
pip install "entroly>=1.0.80"
|
|
40
40
|
openclaw plugins install ./integrations/openclaw
|
|
41
41
|
openclaw plugins enable entroly
|
|
42
42
|
openclaw gateway restart
|
|
@@ -65,8 +65,28 @@ After the first agent turn, run `/entroly-context` in any connected channel to
|
|
|
65
65
|
see the estimated before/after context size, reduction, warnings, and receipt.
|
|
66
66
|
Run `/entroly-context doctor` to verify the configured Python executable and
|
|
67
67
|
local JSONL bridge before inviting users onto the Gateway. The plugin requires
|
|
68
|
-
the
|
|
69
|
-
instead of accepting an older,
|
|
68
|
+
the `entroly.openclaw.bridge.v2` protocol, which shipped in Entroly 1.0.57;
|
|
69
|
+
doctor reports an actionable upgrade instead of accepting an older,
|
|
70
|
+
incompatible Python installation.
|
|
71
|
+
|
|
72
|
+
## Tool output
|
|
73
|
+
|
|
74
|
+
In a long session the tool logs are usually the largest thing in the context: a
|
|
75
|
+
single test, build, or linter run can outweigh every message written around it.
|
|
76
|
+
When the transcript exceeds the budget, Entroly compresses the payload of
|
|
77
|
+
`tool_result` blocks that sit behind the preserved recent turns, using the same
|
|
78
|
+
codecs the proxy applies — specialized codecs first, then the pattern rules,
|
|
79
|
+
then the entropic shell codec — so failures, summaries, and identifiers survive
|
|
80
|
+
while repeated passing lines do not.
|
|
81
|
+
|
|
82
|
+
A block's `tool_use_id` and `is_error` are never rewritten, and any
|
|
83
|
+
`tool_result` carrying a key Entroly does not recognize is passed through
|
|
84
|
+
untouched. Within budget nothing is compressed at all, including tool output.
|
|
85
|
+
The assemble response reports `tool_output_tokens_saved` separately from
|
|
86
|
+
`tokens_saved`, so a receipt reader can tell reduction that came from
|
|
87
|
+
discarding tool noise from reduction that came from compressing what a person
|
|
88
|
+
or the model actually wrote. Set `compress_tool_results: false` on the assemble
|
|
89
|
+
request to disable it.
|
|
70
90
|
|
|
71
91
|
OpenClaw's resolved prompt token budget is authoritative. When an older or
|
|
72
92
|
degraded host cannot provide one, Entroly automatically resolves a conservative
|