Skip to content

Agent Zero AI framework

Restore configurable Vision Model call limits

GreifMax906eee4da715
diff --git a/plugins/_model_config/AGENTS.md b/plugins/_model_config/AGENTS.md
index f6350ea11..1c4dce812 100644
--- a/plugins/_model_config/AGENTS.md
+++ b/plugins/_model_config/AGENTS.md
@@ -26,6 +26,7 @@
- The optional `vision` slot is strictly per preset and never inherited from `Default`; an empty slot disables the separate Vision Model for that preset.
- Main native vision wins by default. A configured Vision Model handles `vision_load` when Main lacks vision, or when that preset explicitly enables `override_main`.
- Keep the optional Vision provider/model selector inside the Main Model card and flush with Main's field alignment, without a nested left inset. Show it only while Main vision is disabled or `override_main` is enabled; do not render a standalone Vision Model card.
+- Keep Vision timeout and maximum-output-token controls, plus the Agent Editor prompt-customization note, inside the visible Vision sidecar's Advanced Settings only. The Vision call limits belong to the preset/model builder, not to `vision_load` call-site constants.
- Show `Use separate Vision Model` immediately below `Supports Vision` while Main vision is enabled, not inside Advanced Settings; describe the disabled state as using Main's native vision.
- In model overviews, render the effective Vision Model as a text-only `Vision override / Provider / Model` child aligned with Main's provider column, not as an icon-bearing peer row.
- Changing a model provider in the settings UI must clear `api_base` and `kwargs` because both may be provider-specific.
diff --git a/plugins/_model_config/helpers/model_config.py b/plugins/_model_config/helpers/model_config.py
index 0c6f539c2..0aa73a37f 100644
--- a/plugins/_model_config/helpers/model_config.py
+++ b/plugins/_model_config/helpers/model_config.py
@@ -11,6 +11,8 @@ PRESETS_FILE = "presets.yaml"
FALLBACK_PRESETS_FILE = "mode_presets_fallback.yaml"
PROVIDER_METADATA_FILE = "provider_metadata.yaml"
DEFAULT_PRESET_NAME = "Default"
+DEFAULT_VISION_TIMEOUT_SECONDS = 300
+DEFAULT_VISION_MAX_TOKENS = 2000
MODEL_PRESET_CONFIG_KEY = "model_preset"
PRESET_SCOPE_GLOBAL = "global"
PRESET_SCOPE_PROJECT = "project"
@@ -25,6 +27,8 @@ IMPLICIT_PRESET_SLOT_DEFAULTS = {
"vision": {
"vision": True,
"max_embeds": 10,
+ "timeout": DEFAULT_VISION_TIMEOUT_SECONDS,
+ "max_tokens": DEFAULT_VISION_MAX_TOKENS,
"override_main": False,
"rl_requests": 0,
"rl_input": 0,
@@ -873,8 +877,18 @@ def build_vision_model(agent=None):
cfg = get_vision_model_config(agent)
mc = build_model_config(cfg, models.ModelType.CHAT)
mc.vision = True
+ kwargs = mc.build_kwargs()
+ for key, default in (
+ ("timeout", DEFAULT_VISION_TIMEOUT_SECONDS),
+ ("max_tokens", DEFAULT_VISION_MAX_TOKENS),
+ ):
+ value = cfg.get(key)
+ if value not in (None, ""):
+ kwargs[key] = _normalize_kwargs({key: value})[key]
+ else:
+ kwargs.setdefault(key, default)
return models.get_chat_model(
- mc.provider, mc.name, model_config=mc, **mc.build_kwargs()
+ mc.provider, mc.name, model_config=mc, **kwargs
)
diff --git a/plugins/_model_config/webui/model-config-store.js b/plugins/_model_config/webui/model-config-store.js
index 1ab6d8c8e..74c723132 100644
--- a/plugins/_model_config/webui/model-config-store.js
+++ b/plugins/_model_config/webui/model-config-store.js
@@ -63,6 +63,8 @@ const IMPLICIT_PRESET_SLOT_DEFAULTS = {
vision: {
vision: true,
max_embeds: 10,
+ timeout: 300,
+ max_tokens: 2000,
override_main: false,
rl_requests: 0,
rl_input: 0,
@@ -222,7 +224,23 @@ export const store = createStore("modelConfig", {
const source = (rawPresets || []).filter(p => p && typeof p === 'object');
const rawDefault = source.find(p => String(p.name || '').toLowerCase() === 'default') || {};
const slot = value => ({ provider: '', name: '', api_key: '', api_base: '', kwargs: {}, ...(value || {}) });
- const visionSlot = value => ({ ...slot(value), vision: true, max_embeds: Number(value?.max_embeds ?? 10), override_main: !!value?.override_main });
+ const visionSlot = value => {
+ const normalized = slot(value);
+ const kwargs = { ...(normalized.kwargs || {}) };
+ const timeout = Number(value?.timeout ?? kwargs.timeout ?? 300);
+ const maxTokens = Number(value?.max_tokens ?? kwargs.max_tokens ?? 2000);
+ delete kwargs.timeout;
+ delete kwargs.max_tokens;
+ return {
+ ...normalized,
+ vision: true,
+ max_embeds: Number(value?.max_embeds ?? 10),
+ timeout,
+ max_tokens: maxTokens,
+ override_main: !!value?.override_main,
+ kwargs,
+ };
+ };
const defaultConfig = {
chat_model: slot(rawDefault.chat),
vision_model: hasModelIdentity(rawDefault.vision) ? visionSlot(rawDefault.vision) : {},
@@ -234,10 +252,11 @@ export const store = createStore("modelConfig", {
const effective = String(p.name || '').toLowerCase() === 'default'
? defaultConfig
: configFromPreset(p, defaultConfig, true);
+ const vision = visionSlot(effective.vision_model);
return {
name: p.name || '',
chat: { ...slot(effective.chat_model), _kwargs_text: kwargsToText(effective.chat_model?.kwargs) },
- vision: { ...visionSlot(effective.vision_model), _kwargs_text: kwargsToText(effective.vision_model?.kwargs) },
+ vision: { ...vision, _kwargs_text: kwargsToText(vision.kwargs) },
utility: { ...slot(effective.utility_model), _kwargs_text: kwargsToText(effective.utility_model?.kwargs) },
embedding: { ...slot(effective.embedding_model), _kwargs_text: kwargsToText(effective.embedding_model?.kwargs) },
};
@@ -360,7 +379,7 @@ export const store = createStore("modelConfig", {
|| this.presets[0]
|| {
chat: { provider: '', name: '', api_base: '', kwargs: {}, _kwargs_text: '' },
- vision: { provider: '', name: '', api_base: '', vision: true, max_embeds: 10, override_main: false, kwargs: {}, _kwargs_text: '' },
+ vision: { provider: '', name: '', api_base: '', vision: true, max_embeds: 10, timeout: 300, max_tokens: 2000, override_main: false, kwargs: {}, _kwargs_text: '' },
utility: { provider: '', name: '', api_base: '', kwargs: {}, _kwargs_text: '' },
embedding: { provider: '', name: '', api_base: '', kwargs: {}, _kwargs_text: '' },
}
diff --git a/plugins/_model_config/webui/model-field.html b/plugins/_model_config/webui/model-field.html
index 51c70741a..a0703ddb3 100644
--- a/plugins/_model_config/webui/model-field.html
+++ b/plugins/_model_config/webui/model-field.html
@@ -9,7 +9,7 @@
<!--
Reusable model configuration field set.
Parent x-data scope must provide:
- model — reactive object with provider, name, api_key, api_base, ctx_length, ctx_history, ctx_input, vision, max_embeds, rl_requests, rl_input, rl_output, kwargs, _kwargs_text
+ model — reactive object with provider, name, api_key, api_base, ctx_length, ctx_history, ctx_input, vision, max_embeds, timeout, max_tokens, rl_requests, rl_input, rl_output, kwargs, _kwargs_text
modelType — 'chat' | 'vision' | 'utility' | 'embedding'
providers — array of { value, label }
searchType — 'chat' | 'embedding'
@@ -224,14 +224,38 @@
</template>
<template x-if="modelType === 'vision'">
- <div class="field">
- <div class="field-label">
- <div class="field-title">Max embeds</div>
- <div class="field-description">Maximum number of images sent to the Vision Model in one call. Set to 0 for unlimited.</div>
+ <div>
+ <div class="field">
+ <div class="field-label">
+ <div class="field-title">Max embeds</div>
+ <div class="field-description">Maximum number of images sent to the Vision Model in one call. Set to 0 for unlimited.</div>
+ </div>
+ <div class="field-control">
+ <input type="number" min="0" x-model.number="model.max_embeds" />
+ </div>
</div>
- <div class="field-control">
- <input type="number" min="0" x-model.number="model.max_embeds" />
+
+ <div class="field">
+ <div class="field-label">
+ <div class="field-title">Timeout (seconds)</div>
+ <div class="field-description">How long to wait for the Vision Model.</div>
+ </div>
+ <div class="field-control">
+ <input type="number" min="1" x-model.number="model.timeout" />
+ </div>
</div>
+
+ <div class="field">
+ <div class="field-label">
+ <div class="field-title">Max tokens</div>
+ <div class="field-description">Maximum output tokens for each delegated vision call.</div>
+ </div>
+ <div class="field-control">
+ <input type="number" min="1" x-model.number="model.max_tokens" />
+ </div>
+ </div>
+
+ <p class="field-description">To customize the vision prompt, use Agent Editor &gt; Advanced &gt; Prompt files and edit <code>fw.vision_load.md</code>.</p>
</div>
</template>
diff --git a/tests/test_model_config_project_presets.py b/tests/test_model_config_project_presets.py
index 1c972b43c..467e57391 100644
--- a/tests/test_model_config_project_presets.py
+++ b/tests/test_model_config_project_presets.py
@@ -1130,6 +1130,104 @@ def test_preset_vision_slot_is_optional_and_never_inherited(monkeypatch, tmp_pat
assert "default-vision" not in str(with_sidecar["vision_model"])
+def test_vision_call_limits_are_preset_owned_and_default_clean(monkeypatch, tmp_path):
+ _prepare_a0_tree(monkeypatch, tmp_path)
+
+ from plugins._model_config.helpers import model_config
+
+ cleaned = model_config.clean_presets_for_file(
+ [
+ {
+ "name": "Default limits",
+ "vision": {
+ "provider": "openrouter",
+ "name": "vision-default",
+ "timeout": 300,
+ "max_tokens": 2000,
+ },
+ },
+ {
+ "name": "Tuned limits",
+ "vision": {
+ "provider": "openrouter",
+ "name": "vision-tuned",
+ "timeout": 45,
+ "max_tokens": 512,
+ },
+ },
+ ]
+ )
+
+ assert cleaned[0]["vision"] == {
+ "provider": "openrouter",
+ "name": "vision-default",
+ }
+ assert cleaned[1]["vision"]["timeout"] == 45
+ assert cleaned[1]["vision"]["max_tokens"] == 512
+
+
+def test_vision_model_build_applies_only_its_preset_call_limits(monkeypatch):
+ from plugins._model_config.helpers import model_config
+
+ calls = []
+
+ def fake_get_chat_model(provider, name, **kwargs):
+ calls.append((provider, name, kwargs))
+ return kwargs
+
+ monkeypatch.setattr(model_config.models, "get_chat_model", fake_get_chat_model)
+
+ cases = [
+ (
+ {"provider": "openrouter", "name": "vision-default"},
+ {"timeout": 300, "max_tokens": 2000},
+ ),
+ (
+ {
+ "provider": "openrouter",
+ "name": "vision-legacy-kwargs",
+ "kwargs": {"timeout": 90, "max_tokens": 1024},
+ },
+ {"timeout": 90, "max_tokens": 1024},
+ ),
+ (
+ {
+ "provider": "openrouter",
+ "name": "vision-tuned",
+ "timeout": "45",
+ "max_tokens": "512",
+ "kwargs": {
+ "timeout": 90,
+ "max_tokens": 1024,
+ "temperature": 0.1,
+ },
+ },
+ {"timeout": 45, "max_tokens": 512},
+ ),
+ ]
+
+ for config, expected in cases:
+ monkeypatch.setattr(
+ model_config,
+ "get_vision_model_config",
+ lambda _agent=None, config=config: config,
+ )
+ built = model_config.build_vision_model()
+ assert built["timeout"] == expected["timeout"]
+ assert built["max_tokens"] == expected["max_tokens"]
+
+ assert calls[-1][2]["temperature"] == 0.1
+
+ monkeypatch.setattr(
+ model_config,
+ "get_chat_model_config",
+ lambda _agent=None: {"provider": "openrouter", "name": "main"},
+ )
+ main = model_config.build_chat_model()
+ assert "timeout" not in main
+ assert "max_tokens" not in main
+
+
def test_legacy_utility_preset_defaults_preserve_tuning_but_clear_kwargs(
monkeypatch,
tmp_path,
diff --git a/tests/test_model_config_ui.py b/tests/test_model_config_ui.py
index 5c6d54aa8..2bc9f72b8 100644
--- a/tests/test_model_config_ui.py
+++ b/tests/test_model_config_ui.py
@@ -98,6 +98,9 @@ def test_preset_editor_nests_one_conditional_vision_selector_in_main() -> None:
override_start = model_field.index('<div class="field-title">Use separate Vision Model</div>')
context_start = model_field.index('<div class="field-title">Context window size</div>')
advanced_start = model_field.index('<!-- Advanced Settings (collapsed by default) -->')
+ vision_advanced_start = model_field.index('<template x-if="modelType === \'vision\'">')
+ vision_advanced_end = model_field.index('<!-- Utility-specific: ctx_input slider -->')
+ vision_advanced = model_field[vision_advanced_start:vision_advanced_end]
assert '<div class="section-title">Vision Model</div>' not in preset_modal
assert preset_modal.count('class="vision-sidecar-selector"') == 1
@@ -115,6 +118,24 @@ def test_preset_editor_nests_one_conditional_vision_selector_in_main() -> None:
assert "When enabled, vision_load uses the preset's Vision Model" not in model_field
assert 'x-model="visionModel.override_main"' in model_field
assert '<div class="advanced-section" x-data="{ advOpen: false }">' in model_field
+ assert advanced_start < vision_advanced_start
+ assert model_field.count('<div class="field-title">Timeout (seconds)</div>') == 1
+ assert model_field.count('<div class="field-title">Max tokens</div>') == 1
+ assert 'x-model.number="model.timeout"' in vision_advanced
+ assert 'x-model.number="model.max_tokens"' in vision_advanced
+ assert "How long to wait for the Vision Model." in vision_advanced
+ assert "Maximum output tokens for each delegated vision call." in vision_advanced
+ assert "Agent Editor &gt; Advanced &gt; Prompt files" in vision_advanced
+ assert "fw.vision_load.md" in vision_advanced
+ assert "delegated_system" not in model_field
+ assert "vision prompt" not in model_field[:vision_advanced_start].lower()
+ assert "timeout: 300" in preset_store
+ assert "max_tokens: 2000" in preset_store
+ assert "value?.timeout ?? kwargs.timeout ?? 300" in preset_store
+ assert "value?.max_tokens ?? kwargs.max_tokens ?? 2000" in preset_store
+ assert "delete kwargs.timeout;" in preset_store
+ assert "delete kwargs.max_tokens;" in preset_store
+ assert "_kwargs_text: kwargsToText(vision.kwargs)" in preset_store
assert "model-preset-row-nested" in preset_overview
assert "model.title === 'Vision'" in preset_overview
assert preset_overview.count("model.title !== 'Vision'") == 2