assistant 2.2.0b1__tar.gz → 2.2.1b1__tar.gz
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.
- {assistant-2.2.0b1/assistant.egg-info → assistant-2.2.1b1}/PKG-INFO +57 -7
- {assistant-2.2.0b1 → assistant-2.2.1b1}/README.md +56 -6
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/__init__.py +5 -1
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/agent.py +102 -26
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/builtin_cmds.py +82 -13
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/config.py +46 -0
- assistant-2.2.1b1/assistant/display.py +134 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/listen/interface.py +6 -7
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/main.py +13 -136
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/memory.py +72 -1
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/ptk_shell/shell.py +504 -119
- assistant-2.2.1b1/assistant/ptk_shell/toolbox.py +132 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/setup_wizard.py +59 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/tools.py +125 -7
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/traces/store.py +93 -40
- {assistant-2.2.0b1 → assistant-2.2.1b1/assistant.egg-info}/PKG-INFO +57 -7
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant.egg-info/SOURCES.txt +5 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/setup.cfg +0 -1
- {assistant-2.2.0b1 → assistant-2.2.1b1}/tests/test_builtin_cmds.py +178 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/tests/test_compaction.py +114 -0
- assistant-2.2.1b1/tests/test_display.py +236 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/tests/test_interrupt.py +17 -25
- assistant-2.2.1b1/tests/test_join_live.py +88 -0
- assistant-2.2.1b1/tests/test_setup_wizard.py +68 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/tests/test_startup_performance.py +8 -5
- assistant-2.2.1b1/tests/test_thinking_loop.py +128 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/tests/test_tools.py +55 -2
- {assistant-2.2.0b1 → assistant-2.2.1b1}/tests/test_tts_streaming.py +199 -25
- assistant-2.2.0b1/tests/test_setup_wizard.py +0 -22
- {assistant-2.2.0b1 → assistant-2.2.1b1}/LICENSE +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/MANIFEST.in +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/__main__.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/as_client.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/as_service.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/codecache.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/compress.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/conditions.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/entry_points/__init__.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/entry_points/run_assistant.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/envindex.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/execer.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/history.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/listen/__init__.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/listen/__main__.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/listen/main.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/nlp/__init__.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/nlp/clear.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/nlp/exit.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/nlp/http_client.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/nlp/interface.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/nlp/message.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/nlp/providers/__init__.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/nlp/providers/google.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/nlp/providers/llama_cpp.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/nlp/providers/nim.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/nlp/providers/ollama.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/nlp/providers/openai.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/nlp/providers/openrouter.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/nlp/providers/vllm.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/nlp/think.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/nlp/xonsh.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/parser.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/projectfile.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/ptk_shell/__init__.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/ptk_shell/bindings.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/ptk_shell/completer.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/ptk_shell/dialogs.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/ptk_shell/floats.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/ptk_shell/layout.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/ptk_shell/operators.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/ptk_shell/strformat_utils.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/ptk_shell/styles.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/ptk_shell/windows.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/say/__init__.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/say/interface.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/time.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/tools_worker.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/traces/__init__.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/traces/cmd.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/traces/export.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/traces/schema.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/utils.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/version.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant/vocabulary.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant.egg-info/dependency_links.txt +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant.egg-info/entry_points.txt +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant.egg-info/requires.txt +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/assistant.egg-info/top_level.txt +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/pyproject.toml +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/setup.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/tests/test_bindings.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/tests/test_compress.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/tests/test_config.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/tests/test_envindex.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/tests/test_envindex_suggestions.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/tests/test_git_hook.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/tests/test_history_bridge.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/tests/test_http_client.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/tests/test_init.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/tests/test_interrupt_real.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/tests/test_listen_interrupt.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/tests/test_memory.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/tests/test_memory_sqlite.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/tests/test_model_compare.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/tests/test_projectfile.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/tests/test_tmux_interactive.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/tests/test_tool_handling.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/tests/test_traces_agent.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/tests/test_traces_cmd.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/tests/test_traces_export.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/tests/test_traces_schema.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/tests/test_traces_store.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/tests/test_utils.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/tests/test_xontrib.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/xontrib/__init__.py +0 -0
- {assistant-2.2.0b1 → assistant-2.2.1b1}/xontrib/assistant.py +0 -0
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
Metadata-Version: 2.4
|
|
2
2
|
Name: assistant
|
|
3
|
-
Version: 2.2.
|
|
3
|
+
Version: 2.2.1b1
|
|
4
4
|
Summary: Your very own Assistant. Because you deserve it.
|
|
5
5
|
Home-page: https://gitlab.com/waser-technologies/technologies/assistant
|
|
6
6
|
Author: Danny Waser
|
|
@@ -126,6 +126,8 @@ pacman -S python-assistant
|
|
|
126
126
|
|
|
127
127
|
</details>
|
|
128
128
|
|
|
129
|
+
Read [Use Assistant as Your Default Shell](#use-assistant-as-your-default-shell) for more information.
|
|
130
|
+
|
|
129
131
|
### 2. Run the Configuration Wizard
|
|
130
132
|
|
|
131
133
|
On the first launch the wizard starts automatically. You can also re-run it at any time:
|
|
@@ -153,7 +155,7 @@ The configuration is saved to `~/.assistant/config.toml`.
|
|
|
153
155
|
assistant
|
|
154
156
|
```
|
|
155
157
|
|
|
156
|
-
That's it
|
|
158
|
+
That's it: you're in. Type commands, ask questions, or mix both.
|
|
157
159
|
|
|
158
160
|
---
|
|
159
161
|
|
|
@@ -434,16 +436,64 @@ Toggle shortcuts (Ctrl+O, Ctrl+Q) update the config automatically when pressed,
|
|
|
434
436
|
|
|
435
437
|
## Use Assistant as Your Default Shell
|
|
436
438
|
|
|
437
|
-
|
|
439
|
+
Assistant is a full Xonsh shell, so it reads the same RC files (`~/.config/xonsh/rc.xsh` or `~/.xonshrc`) which means your existing config carries over as-is.
|
|
440
|
+
|
|
441
|
+
Read more about Xonsh [Installation Guide](https://xon.sh/install.html) and its [Run Control Files](https://xon.sh/xonshrc.html).
|
|
442
|
+
|
|
443
|
+
### Install Assistant in a Stable, Isolated Environment
|
|
444
|
+
|
|
445
|
+
(so a system Python upgrade can't break your shell)
|
|
446
|
+
|
|
447
|
+
**Example with `uv`**
|
|
448
|
+
|
|
449
|
+
```bash
|
|
450
|
+
env_root=~/.venvs
|
|
451
|
+
env_path=${env_root}/assistant
|
|
452
|
+
python_path=${env_path}/bin/python
|
|
453
|
+
mkdir -p $env_root
|
|
454
|
+
uv venv $env_path --python 3.13
|
|
455
|
+
uv pip install --python ${python_path} assistant
|
|
456
|
+
```
|
|
457
|
+
|
|
458
|
+
From there to update Assistant, simply run:
|
|
459
|
+
```bash
|
|
460
|
+
uv pip install --python ${python_path} -U assistant
|
|
461
|
+
```
|
|
462
|
+
|
|
463
|
+
### Create a Dedicated Terminal Profile
|
|
464
|
+
|
|
465
|
+
Since Assistant is a natural language shell and Xonsh is not ready to be fully POSIX compliant,
|
|
466
|
+
it is highly recommended that you create a dedicated profile for Assistant beside Xonsh and your other shells,
|
|
467
|
+
directly inside your terminal application,
|
|
468
|
+
where you can set Assistant as your default shell, instead of registering it for the system.
|
|
469
|
+
|
|
470
|
+
The path to Assistant is defined as shown in this example.
|
|
471
|
+
```bash
|
|
472
|
+
assistant_bin=${env_path}/bin/assistant
|
|
473
|
+
```
|
|
474
|
+
|
|
475
|
+
You can also create a symbolic link to Assistant in `~/.local/bin` (or any other location in `$PATH`) so that Assistant is easily accessible.
|
|
476
|
+
|
|
477
|
+
```bash
|
|
478
|
+
ln -s $assistant_bin ~/.local/bin
|
|
479
|
+
```
|
|
480
|
+
|
|
481
|
+
Or you could add `${env_path}/bin` to your `$PATH`.
|
|
438
482
|
|
|
439
|
-
|
|
483
|
+
### Register Assistant as a Shell for the System
|
|
484
|
+
> [!WARNING]
|
|
485
|
+
> If you read the Xonsh installation guide correctly, you understand that **you shouldn't try this method** unless:
|
|
486
|
+
> 1. you know what you are doing and want to test deeper interaction with the POSIX compliance of the system
|
|
487
|
+
> 2. you are ready to assume the consequences of potentially triggering multiple inference calls every time the system executes a command that exits with a non-zero code.
|
|
488
|
+
>
|
|
489
|
+
> Do the following at your own risk.
|
|
440
490
|
|
|
441
491
|
```bash
|
|
442
|
-
sudo sh -c
|
|
443
|
-
chsh -s $
|
|
492
|
+
sudo sh -c "echo ${assistant_bin} >> /etc/shells"
|
|
493
|
+
chsh -s $assistant_bin
|
|
444
494
|
```
|
|
445
495
|
|
|
446
|
-
Log out and
|
|
496
|
+
Log out and back in, Assistant is now your default shell.
|
|
447
497
|
|
|
448
498
|
---
|
|
449
499
|
|
|
@@ -75,6 +75,8 @@ pacman -S python-assistant
|
|
|
75
75
|
|
|
76
76
|
</details>
|
|
77
77
|
|
|
78
|
+
Read [Use Assistant as Your Default Shell](#use-assistant-as-your-default-shell) for more information.
|
|
79
|
+
|
|
78
80
|
### 2. Run the Configuration Wizard
|
|
79
81
|
|
|
80
82
|
On the first launch the wizard starts automatically. You can also re-run it at any time:
|
|
@@ -102,7 +104,7 @@ The configuration is saved to `~/.assistant/config.toml`.
|
|
|
102
104
|
assistant
|
|
103
105
|
```
|
|
104
106
|
|
|
105
|
-
That's it
|
|
107
|
+
That's it: you're in. Type commands, ask questions, or mix both.
|
|
106
108
|
|
|
107
109
|
---
|
|
108
110
|
|
|
@@ -383,16 +385,64 @@ Toggle shortcuts (Ctrl+O, Ctrl+Q) update the config automatically when pressed,
|
|
|
383
385
|
|
|
384
386
|
## Use Assistant as Your Default Shell
|
|
385
387
|
|
|
386
|
-
|
|
388
|
+
Assistant is a full Xonsh shell, so it reads the same RC files (`~/.config/xonsh/rc.xsh` or `~/.xonshrc`) which means your existing config carries over as-is.
|
|
389
|
+
|
|
390
|
+
Read more about Xonsh [Installation Guide](https://xon.sh/install.html) and its [Run Control Files](https://xon.sh/xonshrc.html).
|
|
391
|
+
|
|
392
|
+
### Install Assistant in a Stable, Isolated Environment
|
|
393
|
+
|
|
394
|
+
(so a system Python upgrade can't break your shell)
|
|
395
|
+
|
|
396
|
+
**Example with `uv`**
|
|
397
|
+
|
|
398
|
+
```bash
|
|
399
|
+
env_root=~/.venvs
|
|
400
|
+
env_path=${env_root}/assistant
|
|
401
|
+
python_path=${env_path}/bin/python
|
|
402
|
+
mkdir -p $env_root
|
|
403
|
+
uv venv $env_path --python 3.13
|
|
404
|
+
uv pip install --python ${python_path} assistant
|
|
405
|
+
```
|
|
406
|
+
|
|
407
|
+
From there to update Assistant, simply run:
|
|
408
|
+
```bash
|
|
409
|
+
uv pip install --python ${python_path} -U assistant
|
|
410
|
+
```
|
|
411
|
+
|
|
412
|
+
### Create a Dedicated Terminal Profile
|
|
413
|
+
|
|
414
|
+
Since Assistant is a natural language shell and Xonsh is not ready to be fully POSIX compliant,
|
|
415
|
+
it is highly recommended that you create a dedicated profile for Assistant beside Xonsh and your other shells,
|
|
416
|
+
directly inside your terminal application,
|
|
417
|
+
where you can set Assistant as your default shell, instead of registering it for the system.
|
|
418
|
+
|
|
419
|
+
The path to Assistant is defined as shown in this example.
|
|
420
|
+
```bash
|
|
421
|
+
assistant_bin=${env_path}/bin/assistant
|
|
422
|
+
```
|
|
423
|
+
|
|
424
|
+
You can also create a symbolic link to Assistant in `~/.local/bin` (or any other location in `$PATH`) so that Assistant is easily accessible.
|
|
425
|
+
|
|
426
|
+
```bash
|
|
427
|
+
ln -s $assistant_bin ~/.local/bin
|
|
428
|
+
```
|
|
429
|
+
|
|
430
|
+
Or you could add `${env_path}/bin` to your `$PATH`.
|
|
387
431
|
|
|
388
|
-
|
|
432
|
+
### Register Assistant as a Shell for the System
|
|
433
|
+
> [!WARNING]
|
|
434
|
+
> If you read the Xonsh installation guide correctly, you understand that **you shouldn't try this method** unless:
|
|
435
|
+
> 1. you know what you are doing and want to test deeper interaction with the POSIX compliance of the system
|
|
436
|
+
> 2. you are ready to assume the consequences of potentially triggering multiple inference calls every time the system executes a command that exits with a non-zero code.
|
|
437
|
+
>
|
|
438
|
+
> Do the following at your own risk.
|
|
389
439
|
|
|
390
440
|
```bash
|
|
391
|
-
sudo sh -c
|
|
392
|
-
chsh -s $
|
|
441
|
+
sudo sh -c "echo ${assistant_bin} >> /etc/shells"
|
|
442
|
+
chsh -s $assistant_bin
|
|
393
443
|
```
|
|
394
444
|
|
|
395
|
-
Log out and
|
|
445
|
+
Log out and back in, Assistant is now your default shell.
|
|
396
446
|
|
|
397
447
|
---
|
|
398
448
|
|
|
@@ -5,7 +5,7 @@ import warnings
|
|
|
5
5
|
warnings.filterwarnings("ignore", category=DeprecationWarning, module="webrtcvad")
|
|
6
6
|
warnings.filterwarnings("ignore", category=UserWarning, module="webrtcvad")
|
|
7
7
|
|
|
8
|
-
__version__ = "2.2.
|
|
8
|
+
__version__ = "2.2.1b1"
|
|
9
9
|
|
|
10
10
|
USERNAME = os.environ.get("USER", "root")
|
|
11
11
|
HOME = os.environ.get("HOME", f"/home/{USERNAME}" if USERNAME != "root" else "/root")
|
|
@@ -131,7 +131,9 @@ DEFAULT_CONFIG = {
|
|
|
131
131
|
"max_turns": 50,
|
|
132
132
|
"temperature": 0.1,
|
|
133
133
|
"max_command_output_length": 10000,
|
|
134
|
+
"max_tool_output_lines": 50,
|
|
134
135
|
"command_timeout": 180,
|
|
136
|
+
"tool_timeouts": {"xonsh": 180, "web_search": 15, "read_webpage": 20},
|
|
135
137
|
},
|
|
136
138
|
"memory": {
|
|
137
139
|
"enabled": False,
|
|
@@ -161,6 +163,8 @@ DEFAULT_CONFIG = {
|
|
|
161
163
|
# end of the system prompt, after the "Relevant commands" block.
|
|
162
164
|
"suggestions_enabled": True,
|
|
163
165
|
"suggestions_budget": 800,
|
|
166
|
+
"thinking_loop_enabled": True,
|
|
167
|
+
"thinking_loop_max_turns": 5,
|
|
164
168
|
},
|
|
165
169
|
"voice": {
|
|
166
170
|
"tts_enabled": True,
|
|
@@ -11,7 +11,7 @@ import types
|
|
|
11
11
|
from typing import Any, Callable, Dict, List, Optional
|
|
12
12
|
|
|
13
13
|
from assistant.config import DEFAULT_CONFIG_PATH, get_config
|
|
14
|
-
from assistant.
|
|
14
|
+
from assistant.nlp.interface import get_random_thinking_sentence
|
|
15
15
|
from assistant.memory import memory
|
|
16
16
|
from assistant.nlp.providers import get_provider
|
|
17
17
|
from assistant.nlp.xonsh import XONSH_INTRO
|
|
@@ -43,6 +43,10 @@ class Agent:
|
|
|
43
43
|
model_cfg = cfg.get_model_config(model_id)
|
|
44
44
|
self.model_name = model_cfg.get("id", "qwen3.5:latest")
|
|
45
45
|
self.context_length = model_cfg.get("context_length", 256000)
|
|
46
|
+
# NOTE: config key "max_token" is shared between the completer (wants 1 = single word)
|
|
47
|
+
# and the main agent (wants a full output budget). The clamp below enforces a minimum
|
|
48
|
+
# of 4096 tokens for the agent, so the completer's max_token=1 is safely overridden.
|
|
49
|
+
self.max_tokens = max(4096, int(model_cfg.get("max_token", 4096)))
|
|
46
50
|
self.think = model_cfg.get("think", False)
|
|
47
51
|
|
|
48
52
|
# Resolve Provider
|
|
@@ -144,6 +148,10 @@ class Orchestrator:
|
|
|
144
148
|
self._summ_model_id = None # Resolved dedicated model id
|
|
145
149
|
self._active_agent = None # Agent of the current chat turn
|
|
146
150
|
|
|
151
|
+
# Thinking-loop detection (Phase 8): consecutive thinking turns + one-shot nudge flag.
|
|
152
|
+
self._consecutive_thinking_turns = 0
|
|
153
|
+
self._thinking_warned = False
|
|
154
|
+
|
|
147
155
|
self._ensure_config()
|
|
148
156
|
|
|
149
157
|
# Session setup: trace recording (Phase 1). Controlled by config
|
|
@@ -661,6 +669,10 @@ class Orchestrator:
|
|
|
661
669
|
total_turns = 0
|
|
662
670
|
max_turns = self._cfg.max_turns or 50
|
|
663
671
|
|
|
672
|
+
# Reset thinking-loop detection state for this chat() call.
|
|
673
|
+
self._consecutive_thinking_turns = 0
|
|
674
|
+
self._thinking_warned = False
|
|
675
|
+
|
|
664
676
|
# Sync the static sys prompt once per session if possible
|
|
665
677
|
# Cache for system prompt tokens to avoid re-calculating identical blocks
|
|
666
678
|
while total_turns < max_turns:
|
|
@@ -774,6 +786,7 @@ class Orchestrator:
|
|
|
774
786
|
temperature=self._cfg.temperature,
|
|
775
787
|
think=use_think,
|
|
776
788
|
tools=use_tools,
|
|
789
|
+
max_tokens=agent.max_tokens,
|
|
777
790
|
)
|
|
778
791
|
if _is_debug:
|
|
779
792
|
sys.stderr.write(f"[DEBUG] Calling API ({agent.model_name})...\n")
|
|
@@ -782,7 +795,7 @@ class Orchestrator:
|
|
|
782
795
|
self._emit(
|
|
783
796
|
{
|
|
784
797
|
"type": "status",
|
|
785
|
-
"text": f"{agent.name}: {get_random_thinking_sentence()
|
|
798
|
+
"text": f"{agent.name}: {get_random_thinking_sentence()}...",
|
|
786
799
|
}
|
|
787
800
|
)
|
|
788
801
|
if _is_debug:
|
|
@@ -823,6 +836,10 @@ class Orchestrator:
|
|
|
823
836
|
return ctx + [{"role": "assistant", "content": ""}]
|
|
824
837
|
raise e
|
|
825
838
|
|
|
839
|
+
# Thinking-loop detection (Phase 8): nudge the model after N consecutive
|
|
840
|
+
# thinking turns so it stops reasoning and answers directly.
|
|
841
|
+
self._detect_thinking_loop(think_text, context_messages)
|
|
842
|
+
|
|
826
843
|
assistant_msg = {"role": "assistant", "content": visible}
|
|
827
844
|
# Count tokens after adding reasoning - reasoning is preserved for trace
|
|
828
845
|
# fidelity (kept in the message), so it counts toward the token budget.
|
|
@@ -1137,6 +1154,42 @@ class Orchestrator:
|
|
|
1137
1154
|
|
|
1138
1155
|
return visible, think_text, tool_calls, is_streamed, response_meta
|
|
1139
1156
|
|
|
1157
|
+
def _detect_thinking_loop(self, think_text, context_messages):
|
|
1158
|
+
"""Detect a thinking loop and nudge the model to answer directly.
|
|
1159
|
+
|
|
1160
|
+
After ``thinking_loop_max_turns`` consecutive turns where the model
|
|
1161
|
+
produced only reasoning (no visible content, no tool calls), inject a
|
|
1162
|
+
one-shot user-role nudge into ``context_messages`` and emit a status
|
|
1163
|
+
event. The nudge fires once per ``chat()`` call (``_thinking_warned``);
|
|
1164
|
+
any turn with visible content or tool calls resets the counter.
|
|
1165
|
+
|
|
1166
|
+
Args:
|
|
1167
|
+
think_text: The reasoning text from this turn (empty if none).
|
|
1168
|
+
context_messages: The mutable message list to append the nudge to.
|
|
1169
|
+
"""
|
|
1170
|
+
if think_text:
|
|
1171
|
+
self._consecutive_thinking_turns += 1
|
|
1172
|
+
loop_max = self._cfg.thinking_loop_max_turns
|
|
1173
|
+
if (
|
|
1174
|
+
self._cfg.thinking_loop_enabled
|
|
1175
|
+
and loop_max > 0
|
|
1176
|
+
and self._consecutive_thinking_turns >= loop_max
|
|
1177
|
+
and not self._thinking_warned
|
|
1178
|
+
):
|
|
1179
|
+
self._thinking_warned = True
|
|
1180
|
+
context_messages.append(
|
|
1181
|
+
{"role": "user", "content": "Please stop reasoning and answer directly."}
|
|
1182
|
+
)
|
|
1183
|
+
self._emit(
|
|
1184
|
+
{
|
|
1185
|
+
"type": "status",
|
|
1186
|
+
"text": "⚠️ detected thinking loop — nudging model to answer directly",
|
|
1187
|
+
}
|
|
1188
|
+
)
|
|
1189
|
+
else:
|
|
1190
|
+
self._consecutive_thinking_turns = 0
|
|
1191
|
+
self._thinking_warned = False
|
|
1192
|
+
|
|
1140
1193
|
def _get_summ_agent(self):
|
|
1141
1194
|
"""Resolve the LLM used by density-layer compaction (couches 1-3).
|
|
1142
1195
|
|
|
@@ -1189,7 +1242,10 @@ class Orchestrator:
|
|
|
1189
1242
|
# Use config limit or a generous 64k default.
|
|
1190
1243
|
# We no longer hard-cap at 32k to allow modern large-context models.
|
|
1191
1244
|
effective_limit = int(agent.context_length or 64000)
|
|
1192
|
-
|
|
1245
|
+
# Reserve a minimum output budget so the model always has room to respond.
|
|
1246
|
+
# The input must never consume more than (context_length - output_reserve).
|
|
1247
|
+
output_reserve = max(4096, int(agent.max_tokens or 4096))
|
|
1248
|
+
context_limit = min(int(effective_limit * threshold), effective_limit - output_reserve)
|
|
1193
1249
|
|
|
1194
1250
|
current_tokens = count_messages_tokens(run_messages)
|
|
1195
1251
|
if current_tokens > context_limit:
|
|
@@ -1212,6 +1268,14 @@ class Orchestrator:
|
|
|
1212
1268
|
run_messages.clear()
|
|
1213
1269
|
run_messages.extend(system_msgs)
|
|
1214
1270
|
run_messages.extend([m for m in context_messages if m["role"] != "system"])
|
|
1271
|
+
# Hard safety net: if compaction still left us over the limit,
|
|
1272
|
+
# drop oldest non-system messages until we fit.
|
|
1273
|
+
while count_messages_tokens(run_messages) > context_limit:
|
|
1274
|
+
non_system = [i for i, m in enumerate(run_messages) if m.get("role") != "system"]
|
|
1275
|
+
if not non_system:
|
|
1276
|
+
break
|
|
1277
|
+
oldest_idx = non_system[0]
|
|
1278
|
+
run_messages.pop(oldest_idx)
|
|
1215
1279
|
# Couche 3 rolling summary: injected as a trailing system block (the
|
|
1216
1280
|
# lead system prompt is never mutated, keeping the KV-cache warm). It
|
|
1217
1281
|
# lives outside the compaction branch so a summary produced earlier is
|
|
@@ -1331,35 +1395,24 @@ class Orchestrator:
|
|
|
1331
1395
|
if "sigint_pipe_w" in _params:
|
|
1332
1396
|
extra_kwargs["sigint_pipe_w"] = sigint_pipe_w
|
|
1333
1397
|
|
|
1398
|
+
_t0 = time.monotonic()
|
|
1334
1399
|
res = fn(args, **extra_kwargs)
|
|
1400
|
+
_wall = time.monotonic() - _t0
|
|
1335
1401
|
except KeyboardInterrupt as ke:
|
|
1336
1402
|
# Re-raise KeyboardInterrupt so parent loop can handle cleanup
|
|
1337
1403
|
raise ke
|
|
1338
1404
|
|
|
1339
1405
|
# Check for abort after tool execution
|
|
1340
1406
|
if self._abort_event.is_set():
|
|
1341
|
-
self._abort_event.clear()
|
|
1342
1407
|
return None
|
|
1343
1408
|
|
|
1344
|
-
|
|
1345
|
-
|
|
1346
|
-
|
|
1347
|
-
|
|
1348
|
-
|
|
1349
|
-
|
|
1350
|
-
|
|
1351
|
-
)
|
|
1352
|
-
|
|
1353
|
-
tool_msg = {
|
|
1354
|
-
"role": "tool",
|
|
1355
|
-
"tool_call_id": t.id,
|
|
1356
|
-
"name": fn_name,
|
|
1357
|
-
"content": res["content"] or "",
|
|
1358
|
-
}
|
|
1359
|
-
# Soft bound (rtk-like): if the output exceeds the context budget,
|
|
1360
|
-
# compress it for the model context (deterministic, command-aware).
|
|
1361
|
-
# The full output is kept on _full_output for trace storage.
|
|
1362
|
-
content = tool_msg["content"]
|
|
1409
|
+
# Structured display payload (e.g. search_conversation rows)
|
|
1410
|
+
# rides the event for the shell; it never enters tool_msg, so
|
|
1411
|
+
# the LLM context stays the flat ``content`` string.
|
|
1412
|
+
# Compute the output ONCE, shared by the model context and the
|
|
1413
|
+
# user display (issue #10: they must always be the same).
|
|
1414
|
+
content = res["content"] or ""
|
|
1415
|
+
display_output = content
|
|
1363
1416
|
if content:
|
|
1364
1417
|
bound = getattr(self._cfg, "max_tool_output_in_context", None)
|
|
1365
1418
|
if not bound:
|
|
@@ -1372,13 +1425,36 @@ class Orchestrator:
|
|
|
1372
1425
|
if bound and len(content) > bound:
|
|
1373
1426
|
from assistant.compress import compress_tool_output
|
|
1374
1427
|
|
|
1375
|
-
|
|
1376
|
-
tool_msg["content"] = compress_tool_output(
|
|
1428
|
+
display_output = compress_tool_output(
|
|
1377
1429
|
content,
|
|
1378
1430
|
command=cmd_display if fn_name == "xonsh" else fn_name,
|
|
1379
1431
|
limit=bound,
|
|
1380
1432
|
)
|
|
1381
|
-
|
|
1433
|
+
|
|
1434
|
+
event = {
|
|
1435
|
+
"type": "tool_output",
|
|
1436
|
+
"command": cmd_display,
|
|
1437
|
+
"tool": fn_name,
|
|
1438
|
+
"output": display_output,
|
|
1439
|
+
"is_error": res["is_error"],
|
|
1440
|
+
"wall": round(_wall, 3),
|
|
1441
|
+
"artifact": len(display_output or ""),
|
|
1442
|
+
}
|
|
1443
|
+
if res.get("results") is not None:
|
|
1444
|
+
event["results"] = res["results"]
|
|
1445
|
+
if res.get("result_kind"):
|
|
1446
|
+
event["result_kind"] = res["result_kind"]
|
|
1447
|
+
self._emit(event)
|
|
1448
|
+
|
|
1449
|
+
tool_msg = {
|
|
1450
|
+
"role": "tool",
|
|
1451
|
+
"tool_call_id": t.id,
|
|
1452
|
+
"name": fn_name,
|
|
1453
|
+
"content": display_output,
|
|
1454
|
+
}
|
|
1455
|
+
if display_output != content:
|
|
1456
|
+
tool_msg["_full_output"] = content
|
|
1457
|
+
tool_msg["_compressed"] = True
|
|
1382
1458
|
# Optimization: populate cache immediately
|
|
1383
1459
|
tool_msg["_tokens"] = count_message_tokens(tool_msg)
|
|
1384
1460
|
context_messages.append(tool_msg)
|
|
@@ -1,12 +1,18 @@
|
|
|
1
1
|
"""Implements a simple echo command for xonsh."""
|
|
2
2
|
|
|
3
3
|
import os
|
|
4
|
+
import sys
|
|
4
5
|
|
|
5
6
|
from xonsh.built_ins import XSH
|
|
6
7
|
from xonsh.tools import unthreadable
|
|
7
8
|
|
|
8
9
|
from assistant.traces.cmd import TRACES_HELP, traces_fn
|
|
9
10
|
|
|
11
|
+
# Byte-identical to what terminfo's `clear` emits for tmux-256color/xterm
|
|
12
|
+
# (`\x1b[H` home + `\x1b[J` erase-below); no scrollback wipe, matching the
|
|
13
|
+
# external binary's behaviour so `clear` still looks like `clear`.
|
|
14
|
+
TERMINAL_CLEAR = "\x1b[H\x1b[J"
|
|
15
|
+
|
|
10
16
|
|
|
11
17
|
def _log_md(shell, content):
|
|
12
18
|
"""Helper to lazily load and log Markdown content."""
|
|
@@ -124,9 +130,40 @@ def pwd_fn(args, stdin=None):
|
|
|
124
130
|
return 0
|
|
125
131
|
|
|
126
132
|
|
|
133
|
+
CLEAR_HELP = """Usage: clear
|
|
134
|
+
Clear the terminal screen.
|
|
135
|
+
|
|
136
|
+
Runs in-process: the reset sequence is written straight to the terminal
|
|
137
|
+
instead of spawning the external `clear` binary through xonsh's command
|
|
138
|
+
pipeline (which costs a fork/exec plus the usual compile and capture pass)."""
|
|
139
|
+
|
|
140
|
+
|
|
141
|
+
def clear_fn(args, stdin=None):
|
|
142
|
+
"""Clear the terminal screen."""
|
|
143
|
+
if args is None:
|
|
144
|
+
args = []
|
|
145
|
+
if "-h" in args or "--help" in args:
|
|
146
|
+
print(CLEAR_HELP)
|
|
147
|
+
return 0
|
|
148
|
+
sys.stdout.write(TERMINAL_CLEAR)
|
|
149
|
+
sys.stdout.flush()
|
|
150
|
+
return 0
|
|
151
|
+
|
|
152
|
+
|
|
127
153
|
def assistant_fn(args, stdin=None):
|
|
128
|
-
|
|
154
|
+
"""Start a new assistant session (no args) or run a direct command.
|
|
155
|
+
|
|
156
|
+
With no arguments the behaviour depends on the context:
|
|
129
157
|
|
|
158
|
+
- interactive: start a brand-new session (reset the conversation context)
|
|
159
|
+
and show the intro exactly like at startup (spinner + streamed panel +
|
|
160
|
+
TTS karaoke).
|
|
161
|
+
- non-interactive (``assistant assistant``): print the intro text in a
|
|
162
|
+
single chunk on stdout, with no streaming display, spinner, or TTS.
|
|
163
|
+
|
|
164
|
+
With a subcommand (``-h``/``version``/...) the matching command runs.
|
|
165
|
+
Any other argument is sent to the assistant as a query.
|
|
166
|
+
"""
|
|
130
167
|
shell = XSH.shell.shell
|
|
131
168
|
|
|
132
169
|
if args is None:
|
|
@@ -136,9 +173,9 @@ def assistant_fn(args, stdin=None):
|
|
|
136
173
|
output = """# Help for the `assistant` command
|
|
137
174
|
```text
|
|
138
175
|
Usage: assistant [SUBCOMMAND]
|
|
139
|
-
|
|
140
|
-
-h, --help
|
|
141
|
-
version
|
|
176
|
+
(no args) New session (interactive) / print the intro (non-interactive)
|
|
177
|
+
-h, --help Show this help
|
|
178
|
+
version Show version
|
|
142
179
|
```
|
|
143
180
|
"""
|
|
144
181
|
_log_md(shell, output)
|
|
@@ -146,19 +183,46 @@ Direct assistant commands:
|
|
|
146
183
|
|
|
147
184
|
if args and args[0] in ("version", "--version", "-V"):
|
|
148
185
|
from assistant import __version__
|
|
186
|
+
# Here (in an interactive session),
|
|
187
|
+
# we should try to feed the string version
|
|
188
|
+
# of Assistant to the LM has a system output
|
|
189
|
+
# like when an error happens and we feed it to the LM
|
|
190
|
+
# with the user input for context.
|
|
191
|
+
# TODO: Implement LM response to version
|
|
192
|
+
_log_md(shell, f"Version {__version__}")
|
|
193
|
+
return 0
|
|
149
194
|
|
|
150
|
-
|
|
195
|
+
if args:
|
|
196
|
+
# A real subcommand/query: send it to the assistant.
|
|
197
|
+
response = shell.kernel_interface.assistant("assistant " + " ".join(args))
|
|
198
|
+
if response:
|
|
199
|
+
_log_md(shell, response)
|
|
151
200
|
return 0
|
|
152
201
|
|
|
153
|
-
|
|
154
|
-
if
|
|
155
|
-
|
|
202
|
+
# No args: new session (interactive) or quick intro (non-interactive).
|
|
203
|
+
if XSH.env.get("XONSH_INTERACTIVE", False):
|
|
204
|
+
try:
|
|
205
|
+
shell.kernel_interface.agent_client.clear_context()
|
|
206
|
+
except Exception:
|
|
207
|
+
pass
|
|
208
|
+
shell.generate_intro()
|
|
209
|
+
else:
|
|
210
|
+
chunks = []
|
|
156
211
|
|
|
157
|
-
|
|
158
|
-
|
|
212
|
+
def _collect(event):
|
|
213
|
+
if event.get("type") == "response_stream":
|
|
214
|
+
c = event.get("text", "")
|
|
215
|
+
if c:
|
|
216
|
+
chunks.append(c)
|
|
159
217
|
|
|
160
|
-
|
|
161
|
-
|
|
218
|
+
try:
|
|
219
|
+
shell.kernel_interface.get_intro(on_event=_collect)
|
|
220
|
+
except Exception as e:
|
|
221
|
+
shell.log_error(e)
|
|
222
|
+
return 0
|
|
223
|
+
intro_text = "".join(chunks).strip()
|
|
224
|
+
if intro_text:
|
|
225
|
+
print(intro_text)
|
|
162
226
|
return 0
|
|
163
227
|
|
|
164
228
|
|
|
@@ -940,7 +1004,7 @@ BUILTIN_COMMANDS = {
|
|
|
940
1004
|
},
|
|
941
1005
|
"assistant": {
|
|
942
1006
|
"fn": assistant_fn,
|
|
943
|
-
"description": "
|
|
1007
|
+
"description": "Start a new session (interactive) or print the intro (non-interactive); -h/version for info",
|
|
944
1008
|
"usage": "Usage: assistant [SUBCOMMAND]",
|
|
945
1009
|
},
|
|
946
1010
|
"listen": {
|
|
@@ -963,4 +1027,9 @@ BUILTIN_COMMANDS = {
|
|
|
963
1027
|
"description": "Print the current working directory",
|
|
964
1028
|
"usage": PWD_HELP,
|
|
965
1029
|
},
|
|
1030
|
+
"clear": {
|
|
1031
|
+
"fn": clear_fn,
|
|
1032
|
+
"description": "Clear the terminal screen",
|
|
1033
|
+
"usage": CLEAR_HELP,
|
|
1034
|
+
},
|
|
966
1035
|
}
|
|
@@ -243,6 +243,27 @@ class Config:
|
|
|
243
243
|
"command_timeout", DEFAULT_CONFIG["execution"]["command_timeout"]
|
|
244
244
|
)
|
|
245
245
|
|
|
246
|
+
@property
|
|
247
|
+
def tool_timeouts(self) -> dict:
|
|
248
|
+
"""Per-tool timeout (seconds) overrides; falls back to command_timeout."""
|
|
249
|
+
return self.raw.get("execution", {}).get(
|
|
250
|
+
"tool_timeouts", DEFAULT_CONFIG["execution"]["tool_timeouts"]
|
|
251
|
+
)
|
|
252
|
+
|
|
253
|
+
def get_tool_timeout(self, tool_name: str) -> int:
|
|
254
|
+
"""Resolve the timeout (seconds) for a tool, falling back to command_timeout.
|
|
255
|
+
|
|
256
|
+
An explicit per-tool value wins; a missing or invalid entry falls back
|
|
257
|
+
to the global ``command_timeout``.
|
|
258
|
+
"""
|
|
259
|
+
timeouts = self.tool_timeouts
|
|
260
|
+
if tool_name in timeouts:
|
|
261
|
+
try:
|
|
262
|
+
return int(timeouts[tool_name])
|
|
263
|
+
except (TypeError, ValueError):
|
|
264
|
+
pass
|
|
265
|
+
return self.command_timeout
|
|
266
|
+
|
|
246
267
|
@property
|
|
247
268
|
def memory_enabled(self) -> bool:
|
|
248
269
|
return self.raw.get("memory", {}).get(
|
|
@@ -288,6 +309,14 @@ class Config:
|
|
|
288
309
|
DEFAULT_CONFIG["execution"]["max_command_output_length"],
|
|
289
310
|
)
|
|
290
311
|
|
|
312
|
+
@property
|
|
313
|
+
def max_tool_output_lines(self) -> int:
|
|
314
|
+
"""Display-only line cap for tool output panels (0 = no line cap)."""
|
|
315
|
+
return self.raw.get("execution", {}).get(
|
|
316
|
+
"max_tool_output_lines",
|
|
317
|
+
DEFAULT_CONFIG["execution"]["max_tool_output_lines"],
|
|
318
|
+
)
|
|
319
|
+
|
|
291
320
|
@property
|
|
292
321
|
def compaction_threshold(self) -> float:
|
|
293
322
|
return self.raw.get("context", {}).get(
|
|
@@ -418,6 +447,23 @@ class Config:
|
|
|
418
447
|
)
|
|
419
448
|
)
|
|
420
449
|
|
|
450
|
+
@property
|
|
451
|
+
def thinking_loop_enabled(self) -> bool:
|
|
452
|
+
"""Toggle for thinking-loop detection (nudge model after N thinking turns)."""
|
|
453
|
+
return self.raw.get("context", {}).get(
|
|
454
|
+
"thinking_loop_enabled", DEFAULT_CONFIG["context"]["thinking_loop_enabled"]
|
|
455
|
+
)
|
|
456
|
+
|
|
457
|
+
@property
|
|
458
|
+
def thinking_loop_max_turns(self) -> int:
|
|
459
|
+
"""Max consecutive thinking turns before a nudge (0 disables)."""
|
|
460
|
+
return int(
|
|
461
|
+
self.raw.get("context", {}).get(
|
|
462
|
+
"thinking_loop_max_turns",
|
|
463
|
+
DEFAULT_CONFIG["context"]["thinking_loop_max_turns"],
|
|
464
|
+
)
|
|
465
|
+
)
|
|
466
|
+
|
|
421
467
|
@property
|
|
422
468
|
def traces_enabled(self) -> bool:
|
|
423
469
|
"""Trace recording switch (Phase 1: on by default, explicit opt-out)."""
|