Sentience agent examples using BU 1.0 & QWen2.5 3B Local LLM #4

rcholic · 2026-01-11T20:59:48Z

.env Example:

BROWSER_USE_API_KEY=bu_browser_use_api_key
SENTIENCE_API_KEY=sk_live_sentience_ai_com_api_key
OPENAI_API_KEY=sk_your_openai_key

Dependencies

pip install browser-use sentienceapi

For local LLM Examples:

pip install browser-use sentienceapi transformers torch accelerate

For local LLM with quantization*

pip install browser-use sentienceapi transformers torch accelerate bitsandbytes

1. Browser-Use LLM with Vision Disabled:

Snapshot Prompt to LLM

Rules: ordinal→DG=1 then ord asc; otherwise imp desc. Use click(ID)/input_text(ID,...).

Elements: ID	role	text	imp	is_primary	docYq	ord	DG	href
49	link	Show HN: Epstein IM – Talk ...	10300	0	0	0	1	epstein
774	textbox		1014	0	6	-	0
238	link	Show HN: AI Vibe Coding Hac...	300	0	2	8	1	devpost
685	link	Show HN: Rocket Launch and ...	300	0	5	27	1	donuttheje
145	link	Show HN: GlyphLang – An AI-...	300	0	1	4	1	ycombinato
355	link	Show HN: VAM Seek – 2D vide...	300	0	3	13	1	github
733	link	Show HN: A website that auc...	300	0	6	29	1	thedailyau
307	link	Show HN: I built an Open So...	300	0	2	11	1	partridge
379	link	Show HN: Vehluna – A visual...	300	0	3	14	1	vercel
403	link	Show HN: Terminal UI for GCP	300	0	3	15	1	github
547	link	Show HN: I built a fast, mi...	300	0	4	21	1	dota2
190	link	Show HN: Porting xv6 to HiF...	300	0	1	6	1	github
523	link	Show HN: Various shape regu...	300	0	4	20	1	github
709	link	Show HN: EuConform – Offlin...	300	0	5	28	1	github
121	link	Show HN: I used Claude Code...	300	0	1	3	1	pieterma
214	link	Show HN: Play poker with LL...	300	0	2	7	1	llmholdem
619	link	Show HN: Executable Markdow...	300	0	5	24	1	ycombinato
499	link	Show HN: I made a memory ga...	300	0	4	19	1	specr
262	link	Show HN: Librario, a book m...	300	0	2	9	1	ycombinato
97	link	Show HN: A MCP for controll...	300	0	1	2	1	github
571	link	Show HN: We Identified a Co...	300	0	4	22	1	wordpress
331	link	Show HN: Yellopages – New t...	300	0	3	12	1	kawaicheun
73	link	Show HN: Interactive Califo...	300	0	1	1	1	california
427	link	Show HN: Marten – Elegant G...	300	0	3	16	1	github
451	link	Show HN: Enforcing time-bou...	300	0	3	17	1	github
595	link	Show HN: ESPTimeCast – DIY ...	300	0	4	23	1	github
640	link	Show HN: I put economic rul...	300	0	5	25	1	ycombinato
283	link	Show HN: mcpc – Universal c...	300	0	2	10	1	github
166	link	Show HN: Ferrite – Markdown...	300	0	1	5	1	github
661	link	Show HN: Gyesme – exploring...	300	0	5	26	1	gyesme
475	link	Show HN: Umaro – An interac...	300	0	4	18	1	umaro
16	link	Hacker News	0	1	0	-	0	ycombinato
22	link	show	0	0	0	-	0	ycombinato
17	link	new	0	0	0	-	0	ycombinato
11	link		0	0	0	-	0	ycombinato
12	link		0	0	0	-	0	ycombinato
20	link	ask	0	0	0	-	0	ycombinato
18	link	past	0	0	0	-	0	ycombinato
19	link	comments	0	0	0	-	0	ycombinato

INFO     [agent] ✅ Using Sentience snapshot for prompt
INFO     [agent] Context prepared: sentience_used=True, message_length=3621 chars (~905 tokens), history_length=807 chars (~201 tokens)
INFO     [agent] 🤖 Calling LLM (this may trigger model download/loading for Hugging Face models)...
INFO     [agent] ✅ LLM response received
INFO     [agent] LLM response received: 1 actions
INFO     [agent]   ▶️  done: {'text': 'Top post: element ID 49, title: Show HN: Epstein IM – Talk ...', 'success': True}
INFO     [agent]   ✅ Task done: Task completed: True - Top post: element ID 49, title: Show HN: Epstein IM – Talk ...
INFO     [message_manager] Updated history: step=1, items_tracked=3, items_sent_to_llm=3
INFO     [agent] ✅ Task completed
INFO     [agent] Agent completed: total_prompt_tokens=14281 total_prompt_cost=0.0028562

Per Model Usage

Model	Prompt Tokens	Completion Tokens	Total Tokens	Cost	Invocations	Avg Tokens/Invocation
bu-1-0	14,281	343	14,624	$0.0035422	2	7,312.0

📊 Token Usage Summary:
Total tokens: 14624
Total cost: $0.003542
Steps: 2
Sentience used: True
Sentience usage: 2/2 steps (100.0%)

2. QWen2.5 3B Local LLM:

Snapshot Prompt to LLM

Rules: ordinal→DG=1 then ord asc; otherwise imp desc. Use click(ID)/input_text(ID,...).

Elements ID	role	text	imp	is_primary	docYq	ord	DG	href
49	link	Show HN: Epstein IM – Talk ...	10300	0	0	0	1	epstein
774	textbox		1014	0	6	-	0
238	link	Show HN: AI Vibe Coding Hac...	300	0	2	8	1	devpost
685	link	Show HN: Rocket Launch and ...	300	0	5	27	1	donuttheje
145	link	Show HN: GlyphLang – An AI-...	300	0	1	4	1	ycombinato
355	link	Show HN: VAM Seek – 2D vide...	300	0	3	13	1	github
733	link	Show HN: A website that auc...	300	0	6	29	1	thedailyau
307	link	Show HN: I built an Open So...	300	0	2	11	1	partridge
379	link	Show HN: Vehluna – A visual...	300	0	3	14	1	vercel
403	link	Show HN: Terminal UI for GCP	300	0	3	15	1	github
547	link	Show HN: I built a fast, mi...	300	0	4	21	1	dota2
190	link	Show HN: Porting xv6 to HiF...	300	0	1	6	1	github
523	link	Show HN: Various shape regu...	300	0	4	20	1	github
709	link	Show HN: EuConform – Offlin...	300	0	5	28	1	github
121	link	Show HN: I used Claude Code...	300	0	1	3	1	pieterma
214	link	Show HN: Play poker with LL...	300	0	2	7	1	llmholdem
619	link	Show HN: Executable Markdow...	300	0	5	24	1	ycombinato
499	link	Show HN: I made a memory ga...	300	0	4	19	1	specr
262	link	Show HN: Librario, a book m...	300	0	2	9	1	ycombinato
97	link	Show HN: A MCP for controll...	300	0	1	2	1	github
571	link	Show HN: We Identified a Co...	300	0	4	22	1	wordpress
331	link	Show HN: Yellopages – New t...	300	0	3	12	1	kawaicheun
73	link	Show HN: Interactive Califo...	300	0	1	1	1	california
427	link	Show HN: Marten – Elegant G...	300	0	3	16	1	github
451	link	Show HN: Enforcing time-bou...	300	0	3	17	1	github
595	link	Show HN: ESPTimeCast – DIY ...	300	0	4	23	1	github
640	link	Show HN: I put economic rul...	300	0	5	25	1	ycombinato
283	link	Show HN: mcpc – Universal c...	300	0	2	10	1	github
166	link	Show HN: Ferrite – Markdown...	300	0	1	5	1	github
661	link	Show HN: Gyesme – exploring...	300	0	5	26	1	gyesme
475	link	Show HN: Umaro – An interac...	300	0	4	18	1	umaro
16	link	Hacker News	0	1	0	-	0	ycombinato
22	link	show	0	0	0	-	0	ycombinato
17	link	new	0	0	0	-	0	ycombinato
11	link		0	0	0	-	0	ycombinato
12	link		0	0	0	-	0	ycombinato
20	link	ask	0	0	0	-	0	ycombinato
18	link	past	0	0	0	-	0	ycombinato
19	link	comments	0	0	0	-	0	ycombinato

INFO     [agent] ✅ Using Sentience snapshot for prompt
INFO     [agent] Context prepared: sentience_used=True, message_length=3262 chars (~815 tokens), history_length=448 chars (~112 tokens)
INFO     [agent] 🤖 Calling LLM (this may trigger model download/loading for Hugging Face models)...
INFO     [agent] ✅ LLM response received
INFO     [agent] LLM response received: 1 actions
INFO     [agent]   ▶️  done: {'text': 'Top post: element ID 49, title: Show HN: Epstein IM – Talk ...', 'files_to_display': []}
INFO     [agent]   ✅ Task done: Task completed: True - Top post: element ID 49, title: Show HN: Epstein IM – Talk ...
INFO     [message_manager] Updated history: step=0, items_tracked=2, items_sent_to_llm=2
INFO     [agent] ✅ Task completed
INFO     [agent] Agent completed: total_prompt_tokens=0 total_prompt_cost=0.0 total_prompt_cached_tokens=0 total_prompt_cached_cost=0.0 total_completion_tokens=0 total_completion_cost=0.0 total_tokens=0 total_cost=0.0 entry_count=0 by_model={}

Per Model Usage

Model	Prompt Tokens	Completion Tokens	Total Tokens	Cost	Invocations	Avg Tokens/Invocation
Qwen/Qwen2.5-3B-Instruct	5,578	248	5,826	$0.00	1	5,826.0

📊 Token Usage Summary:
Total tokens: 5826
Total cost: $0.000000
Steps: 1
Sentience used: True
Sentience usage: 1/1 steps (100.0%)

INFO     [BrowserSession] 📢 on_BrowserStopEvent - Calling reset() (force=False, keep_alive=None)
INFO     [BrowserSession] [SessionManager] Cleared all owned data (targets, sessions, mappings)
INFO     [BrowserSession] ✅ Browser session reset complete
INFO     [BrowserSession] ✅ Browser session reset complete

When to Use Quantization?

Use if: You have limited RAM/VRAM, want faster inference, or need to run larger models on consumer hardware
Skip if: You have plenty of memory, want maximum quality, or are running on high-end GPUs

SentienceDEV added 8 commits January 10, 2026 22:34

Phase 1 and 2

d8d7fbc

phase 3 done

6aebfe4

debugging

765c194

consistent success

9a51d7b

working agent with less tokens

f963da6

working examples

fd1212c

example agents with bu LLM model and QWen 2.5 3B

705d9ab

updated token usage

3f41fe4

rcholic changed the title ~~Sentience agent~~ Sentience agent examples using BU 1.0 & QWen2.5 3B Local LLM Jan 11, 2026

vision llm as fallback

0a5d1b9

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Sentience agent examples using BU 1.0 & QWen2.5 3B Local LLM #4

Sentience agent examples using BU 1.0 & QWen2.5 3B Local LLM #4

Uh oh!

rcholic commented Jan 11, 2026 •

edited

Loading

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Sentience agent examples using BU 1.0 & QWen2.5 3B Local LLM #4

Are you sure you want to change the base?

Sentience agent examples using BU 1.0 & QWen2.5 3B Local LLM #4

Uh oh!

Conversation

rcholic commented Jan 11, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

.env Example:

Dependencies

For local LLM Examples:

For local LLM with quantization*

1. Browser-Use LLM with Vision Disabled:

Snapshot Prompt to LLM

Per Model Usage

2. QWen2.5 3B Local LLM:

Snapshot Prompt to LLM

Per Model Usage

When to Use Quantization?

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

rcholic commented Jan 11, 2026 •

edited

Loading