Pinned Tweet
It is becoming f*cking dangerous
A 23-year-old dev from China just ran 300 Claude Code agents in parallel and hooked them to a verification layer so none of them could bullshit him
He opens the live dashboard:
- 300 agents writing and checking at the same time
- Every output urce in ~100 ms
- Three passes over 100 company profiles → zero errors left
- Everything dumped straight into an Obsidian vault
The trick is the split: one model does the work, a cheaper one decides if the claim is actually true. The loop doesn’t stop until everything checks out
An analyst team would bill weeks for this. He got it in one run
Same kind of Second Brain + swarm setup that keeps circulating. Bookmark the full breakdown
A 23-year-old dev from China just ran 300 Claude Code agents in parallel and hooked them to a verification layer so none of them could bullshit him
He opens the live dashboard:
- 300 agents writing and checking at the same time
- Every output urce in ~100 ms
- Three passes over 100 company profiles → zero errors left
- Everything dumped straight into an Obsidian vault
The trick is the split: one model does the work, a cheaper one decides if the claim is actually true. The loop doesn’t stop until everything checks out
An analyst team would bill weeks for this. He got it in one run
Same kind of Second Brain + swarm setup that keeps circulating. Bookmark the full breakdown
715
1,017
13,583
6,966,834
Lyren_o retweeted
Apple shows the new MacBook Pro on Oct 28, and someone already rebuilt the rumored version in code with Claude
126 parts, a working lid, fans and heat pipes, all as a 3D model. That's genuinely impressive
What I like most is the method. Every part gets checked before the next one goes in, so the AI catches its own mistakes along the way
Worth remembering that the OLED touchscreen and Dynamic Island are still rumors, so this is a smart guess, not the final design. That's actually what makes it fun
A model like this used to take a designer weeks. Now it's a good prompt and some patience
I'm saving it and will compare it with the real thing on Oct 28
126 parts, a working lid, fans and heat pipes, all as a 3D model. That's genuinely impressive
What I like most is the method. Every part gets checked before the next one goes in, so the AI catches its own mistakes along the way
Worth remembering that the OLED touchscreen and Dynamic Island are still rumors, so this is a smart guess, not the final design. That's actually what makes it fun
A model like this used to take a designer weeks. Now it's a good prompt and some patience
I'm saving it and will compare it with the real thing on Oct 28
THIS IS F*CKING INSANE
Apple shows the new MacBook Pro on Oct 28. Opus 5.5 rebuilt the leaked one first: all 126 parts, and the lid opens.
335k triangles, 259 design rules, 0 errors left.
The trick: every part is code, and nothing goes in until its check passes.
what's inside
en with the Dynamic Island cut-out
> M5 Pro / M5 Max logic board, unified memory on the chip
> 6 battery cells under the keyboard and palm rest
> 2 fans, a vapor plate and 2 heat pipes over the chip
> six speakers, MagSafe, keyboard, trackpad
> the hinge, with the display cables running through it
the checks that gate every part
→ clearance ≥ 0.05 mm, 0 clashes. tightest spot: the lid sits 0.14 mm over the keys
→ the fans get a clear air path past the heat pipes
→ display cables bend ≥ 0.8 mm in the hinge (these bend at 4 mm). that cable is what killed 2016 MacBook screens (flexgate)
→ 259 design rules pass, then the lid opens live to 112°
it caught a real one: the first heat pipe route blocked the fan outlet by 0.60 mm. it moved the pipe 1.5 mm and the air path cleared by 0.90 mm.
The guide I'm quoting rebuilds software with REA and calls this check the oracle. Same loop, pointed at a laptop Apple hasn't even shown yet.
paste this into Claude Code ↓
"Rebuild [any device] from its leaks, in code:
1. List the specs. Mark each one confirmed or rumored.
2. Write the parts list with dimensions. One function per part.
3. Place parts one by one. After each, check clearance, clashes, cable bends and airflow. Fix every fail before the next part.
4. Animate the moving parts last.
Show me the specs and the parts list first. No code until I say go."
Works on anything with a hinge or a fan: consoles, cameras, drones.
save it for Oct 28.
5
1
19
477
Apple shows the new MacBook Pro on Oct 28, and someone already rebuilt the rumored version in code with Claude
126 parts, a working lid, fans and heat pipes, all as a 3D model. That's genuinely impressive
What I like most is the method. Every part gets checked before the next one goes in, so the AI catches its own mistakes along the way
Worth remembering that the OLED touchscreen and Dynamic Island are still rumors, so this is a smart guess, not the final design. That's actually what makes it fun
A model like this used to take a designer weeks. Now it's a good prompt and some patience
I'm saving it and will compare it with the real thing on Oct 28
126 parts, a working lid, fans and heat pipes, all as a 3D model. That's genuinely impressive
What I like most is the method. Every part gets checked before the next one goes in, so the AI catches its own mistakes along the way
Worth remembering that the OLED touchscreen and Dynamic Island are still rumors, so this is a smart guess, not the final design. That's actually what makes it fun
A model like this used to take a designer weeks. Now it's a good prompt and some patience
I'm saving it and will compare it with the real thing on Oct 28
THIS IS F*CKING INSANE
Apple shows the new MacBook Pro on Oct 28. Opus 5.5 rebuilt the leaked one first: all 126 parts, and the lid opens.
335k triangles, 259 design rules, 0 errors left.
The trick: every part is code, and nothing goes in until its check passes.
what's inside
en with the Dynamic Island cut-out
> M5 Pro / M5 Max logic board, unified memory on the chip
> 6 battery cells under the keyboard and palm rest
> 2 fans, a vapor plate and 2 heat pipes over the chip
> six speakers, MagSafe, keyboard, trackpad
> the hinge, with the display cables running through it
the checks that gate every part
→ clearance ≥ 0.05 mm, 0 clashes. tightest spot: the lid sits 0.14 mm over the keys
→ the fans get a clear air path past the heat pipes
→ display cables bend ≥ 0.8 mm in the hinge (these bend at 4 mm). that cable is what killed 2016 MacBook screens (flexgate)
→ 259 design rules pass, then the lid opens live to 112°
it caught a real one: the first heat pipe route blocked the fan outlet by 0.60 mm. it moved the pipe 1.5 mm and the air path cleared by 0.90 mm.
The guide I'm quoting rebuilds software with REA and calls this check the oracle. Same loop, pointed at a laptop Apple hasn't even shown yet.
paste this into Claude Code ↓
"Rebuild [any device] from its leaks, in code:
1. List the specs. Mark each one confirmed or rumored.
2. Write the parts list with dimensions. One function per part.
3. Place parts one by one. After each, check clearance, clashes, cable bends and airflow. Fix every fail before the next part.
4. Animate the moving parts last.
Show me the specs and the parts list first. No code until I say go."
Works on anything with a hinge or a fan: consoles, cameras, drones.
save it for Oct 28.
5
1
19
477
Lyren_o retweeted
ceo who pays 200 people to sit in an office after visiting the tavus office and realizing PALs dont need salaries, coffee breaks or sick days
Tavus didn't announce an office. They let people walk into it.
A 3D space in the browser, a live PAL at every desk, ready to talk face to face. A Corvette that turns into a racing game.
The Tavus team is an excellent example of the right approach to advertising and engaging d to be working with them.
9
1
30
544
ceo who pays 200 people to sit in an office after visiting the tavus office and realizing PALs dont need salaries, coffee breaks or sick days
Tavus didn't announce an office. They let people walk into it.
A 3D space in the browser, a live PAL at every desk, ready to talk face to face. A Corvette that turns into a racing game.
The Tavus team is an excellent example of the right approach to advertising and engaging d to be working with them.
9
1
30
544
Lyren_o retweeted
ok this one got personal. a trebuchet has one job, the rock hits the tower or it doesnt
same prompt, four models, zero edits from me
Claude Opus 5.5 vs Gemini 4 Argon vs GPT-6 Astra vs Grok 4.7
counterweight, sling, release timing. this is where fake physics gets exposed real winner. you tell me
which one actually breaks the tower and which one is just throwing rocks at the sky?
same prompt, four models, zero edits from me
Claude Opus 5.5 vs Gemini 4 Argon vs GPT-6 Astra vs Grok 4.7
counterweight, sling, release timing. this is where fake physics gets exposed real winner. you tell me
which one actually breaks the tower and which one is just throwing rocks at the sky?
26
6
85
9,645
ok this one got personal. a trebuchet has one job, the rock hits the tower or it doesnt
same prompt, four models, zero edits from me
Claude Opus 5.5 vs Gemini 4 Argon vs GPT-6 Astra vs Grok 4.7
counterweight, sling, release timing. this is where fake physics gets exposed real winner. you tell me
which one actually breaks the tower and which one is just throwing rocks at the sky?
same prompt, four models, zero edits from me
Claude Opus 5.5 vs Gemini 4 Argon vs GPT-6 Astra vs Grok 4.7
counterweight, sling, release timing. this is where fake physics gets exposed real winner. you tell me
which one actually breaks the tower and which one is just throwing rocks at the sky?
26
6
85
9,645
Lyren_o retweeted
Most people burn Claude and GPT limits on tasks that do not need them
I added a simple router so cheap models handle the routine work and the strong model is called only when it actually matters
How it works:
1. Router (router.py)
Looks at the task and picks the formatting: cheap model
Needs reasoning or new content: strong model
Pure action: code, no model at all
2. Decision file (decisions.md)
Short rules the router follows.
If the answer already exists in the vault, do not call any model.
If confidence is high and the task is repetitive, use the cheap model.
Only escalate when the output must be original.
3. Execution
Code runs tools and writes files.
The strong model is called only for the final draft or complex synthesis.
Result: most daily tasks stay on the cheap path. The expensive model is used only when it changes the output.
I added a simple router so cheap models handle the routine work and the strong model is called only when it actually matters
How it works:
1. Router (router.py)
Looks at the task and picks the formatting: cheap model
Needs reasoning or new content: strong model
Pure action: code, no model at all
2. Decision file (decisions.md)
Short rules the router follows.
If the answer already exists in the vault, do not call any model.
If confidence is high and the task is repetitive, use the cheap model.
Only escalate when the output must be original.
3. Execution
Code runs tools and writes files.
The strong model is called only for the final draft or complex synthesis.
Result: most daily tasks stay on the cheap path. The expensive model is used only when it changes the output.
22
1
50
1,056
Most people burn Claude and GPT limits on tasks that do not need them
I added a simple router so cheap models handle the routine work and the strong model is called only when it actually matters
How it works:
1. Router (router.py)
Looks at the task and picks the formatting: cheap model
Needs reasoning or new content: strong model
Pure action: code, no model at all
2. Decision file (decisions.md)
Short rules the router follows.
If the answer already exists in the vault, do not call any model.
If confidence is high and the task is repetitive, use the cheap model.
Only escalate when the output must be original.
3. Execution
Code runs tools and writes files.
The strong model is called only for the final draft or complex synthesis.
Result: most daily tasks stay on the cheap path. The expensive model is used only when it changes the output.
I added a simple router so cheap models handle the routine work and the strong model is called only when it actually matters
How it works:
1. Router (router.py)
Looks at the task and picks the formatting: cheap model
Needs reasoning or new content: strong model
Pure action: code, no model at all
2. Decision file (decisions.md)
Short rules the router follows.
If the answer already exists in the vault, do not call any model.
If confidence is high and the task is repetitive, use the cheap model.
Only escalate when the output must be original.
3. Execution
Code runs tools and writes files.
The strong model is called only for the final draft or complex synthesis.
Result: most daily tasks stay on the cheap path. The expensive model is used only when it changes the output.
22
1
50
1,056
Lyren_o retweeted
I WALKED INTO THE OFFICE AND STARTED TALKING TO COLLEAGUES WHO DON’T ACTUALLY EXIST
I went through the front desk, stopped at a few tables, and just started talking
No script, no pre-written answers - a live conversation in real time with people Claude built from a single tion, Tim in partnerships… they’re not just standing there smiling
They reply, ask follow-ups, and keep the conversation going
At some point you start forgetting they’re not real people
While everyone else is yelling “wild” and “insane,” I just checked whether this actually works as a workplace. It does
And that feels more unsettling than any metaverse from 2021–2022
Video inside
I went through the front desk, stopped at a few tables, and just started talking
No script, no pre-written answers - a live conversation in real time with people Claude built from a single tion, Tim in partnerships… they’re not just standing there smiling
They reply, ask follow-ups, and keep the conversation going
At some point you start forgetting they’re not real people
While everyone else is yelling “wild” and “insane,” I just checked whether this actually works as a workplace. It does
And that feels more unsettling than any metaverse from 2021–2022
Video inside
I asked Claude to build an office for the @tavus PALs...
I was not expecting it to build THAT kind of office
15
1
45
914
I WALKED INTO THE OFFICE AND STARTED TALKING TO COLLEAGUES WHO DON’T ACTUALLY EXIST
I went through the front desk, stopped at a few tables, and just started talking
No script, no pre-written answers - a live conversation in real time with people Claude built from a single tion, Tim in partnerships… they’re not just standing there smiling
They reply, ask follow-ups, and keep the conversation going
At some point you start forgetting they’re not real people
While everyone else is yelling “wild” and “insane,” I just checked whether this actually works as a workplace. It does
And that feels more unsettling than any metaverse from 2021–2022
Video inside
I went through the front desk, stopped at a few tables, and just started talking
No script, no pre-written answers - a live conversation in real time with people Claude built from a single tion, Tim in partnerships… they’re not just standing there smiling
They reply, ask follow-ups, and keep the conversation going
At some point you start forgetting they’re not real people
While everyone else is yelling “wild” and “insane,” I just checked whether this actually works as a workplace. It does
And that feels more unsettling than any metaverse from 2021–2022
Video inside
I asked Claude to build an office for the @tavus PALs...
I was not expecting it to build THAT kind of office
15
1
45
914
Lyren_o retweeted
RAILWAY JUST INVALIDATED EVERY API TOKEN AND PAUSED BUILDS. AGAIN
If you’re running agent sandboxes there, the chart is hard to ignore: 10 small sandboxes 24/7 = $118/mo on Lizard vs ~$2,129 on Railway. That’s roughly 18x more expensive before you even count the downtime
I told Lizard”. It handled the services, env vars and domains. First thing I did was rotate every secret. First week on Lizard is free with $5 credits
@LizardBuild
Link: lizard.build/bot
If you’re running agent sandboxes there, the chart is hard to ignore: 10 small sandboxes 24/7 = $118/mo on Lizard vs ~$2,129 on Railway. That’s roughly 18x more expensive before you even count the downtime
I told Lizard”. It handled the services, env vars and domains. First thing I did was rotate every secret. First week on Lizard is free with $5 credits
@LizardBuild
Link: lizard.build/bot
16
2
40
746
RAILWAY JUST INVALIDATED EVERY API TOKEN AND PAUSED BUILDS. AGAIN
If you’re running agent sandboxes there, the chart is hard to ignore: 10 small sandboxes 24/7 = $118/mo on Lizard vs ~$2,129 on Railway. That’s roughly 18x more expensive before you even count the downtime
I told Lizard”. It handled the services, env vars and domains. First thing I did was rotate every secret. First week on Lizard is free with $5 credits
@LizardBuild
Link: lizard.build/bot
If you’re running agent sandboxes there, the chart is hard to ignore: 10 small sandboxes 24/7 = $118/mo on Lizard vs ~$2,129 on Railway. That’s roughly 18x more expensive before you even count the downtime
I told Lizard”. It handled the services, env vars and domains. First thing I did was rotate every secret. First week on Lizard is free with $5 credits
@LizardBuild
Link: lizard.build/bot
Builds are temporarily paused
We are currently investigating this issue
Status: Investigating status.railway.com/incident/QH8L8…
16
2
40
746
Lyren_o retweeted
How I reduced agent token waste by fixing the loop
Most people focus on the model. The bigger leak is usually in the loop around it
I broke my setup into three layers and cleaned each one:
1. Wrapper
Persistent context instead of rebuilding the prompt every call. Fixed hanges. Cuts repeated system instructions
2. Tool layer
Parallel tool calls where possible and strict schemas so the model doesn’t re-explain itself. Cache results of pure lookups
3. Inference
Route simple decisions to cheaper/faster models and keep the strong one only for the final reasoning step. Speculative checks on short outputs
Main rule I follow now: never recompute the same context twice in one task
The terminal video shows the before/after difference on the same request
Most people focus on the model. The bigger leak is usually in the loop around it
I broke my setup into three layers and cleaned each one:
1. Wrapper
Persistent context instead of rebuilding the prompt every call. Fixed hanges. Cuts repeated system instructions
2. Tool layer
Parallel tool calls where possible and strict schemas so the model doesn’t re-explain itself. Cache results of pure lookups
3. Inference
Route simple decisions to cheaper/faster models and keep the strong one only for the final reasoning step. Speculative checks on short outputs
Main rule I follow now: never recompute the same context twice in one task
The terminal video shows the before/after difference on the same request
14
6
43
1,050
How I reduced agent token waste by fixing the loop
Most people focus on the model. The bigger leak is usually in the loop around it
I broke my setup into three layers and cleaned each one:
1. Wrapper
Persistent context instead of rebuilding the prompt every call. Fixed hanges. Cuts repeated system instructions
2. Tool layer
Parallel tool calls where possible and strict schemas so the model doesn’t re-explain itself. Cache results of pure lookups
3. Inference
Route simple decisions to cheaper/faster models and keep the strong one only for the final reasoning step. Speculative checks on short outputs
Main rule I follow now: never recompute the same context twice in one task
The terminal video shows the before/after difference on the same request
Most people focus on the model. The bigger leak is usually in the loop around it
I broke my setup into three layers and cleaned each one:
1. Wrapper
Persistent context instead of rebuilding the prompt every call. Fixed hanges. Cuts repeated system instructions
2. Tool layer
Parallel tool calls where possible and strict schemas so the model doesn’t re-explain itself. Cache results of pure lookups
3. Inference
Route simple decisions to cheaper/faster models and keep the strong one only for the final reasoning step. Speculative checks on short outputs
Main rule I follow now: never recompute the same context twice in one task
The terminal video shows the before/after difference on the same request
14
6
43
1,050
Lyren_o retweeted
I hit a flaky integration test that kept failing only on CI
Instead of debugging it myself I took a snapshot of one Lizard sandbox, forked it into 50 copies in under a second each, and let every Claude Code try a different fix
The first one that got the tests green won
The and cost nothing
Total runtime on the winning path was a few minutes. I never opened the code myself
@LizardBuild makes the snapshot → fork path cheap enough that throwing agents at the same bug stops being a waste of money
Pro is $19/mo with $19 credits inside, first week free
Link: lizard.build/snapshot
Instead of debugging it myself I took a snapshot of one Lizard sandbox, forked it into 50 copies in under a second each, and let every Claude Code try a different fix
The first one that got the tests green won
The and cost nothing
Total runtime on the winning path was a few minutes. I never opened the code myself
@LizardBuild makes the snapshot → fork path cheap enough that throwing agents at the same bug stops being a waste of money
Pro is $19/mo with $19 credits inside, first week free
Link: lizard.build/snapshot
I pointed 100 sandboxes on @LizardBuild at the same messy repo and told each Claude Code to find and fix one class of bugs
One looked for missing error handling, another for flaky tests, another for dead code, another for unsafe SQL. They all started at the same time
Median batch was $1.62/h while they ran. As soon as an agent finished its pass and opened a clean PR, I paused that sandbox and it dropped to $0
By the end of the hour I had 73 PRs sitting ready, each with a different fix and a short note on what it changed. I didn’t babysit a single process
Pro is $19/mo with $19 credits inside. First week free with $5 to try
Link: lizard.build/sandboxes/code
10
2
30
913
I hit a flaky integration test that kept failing only on CI
Instead of debugging it myself I took a snapshot of one Lizard sandbox, forked it into 50 copies in under a second each, and let every Claude Code try a different fix
The first one that got the tests green won
The and cost nothing
Total runtime on the winning path was a few minutes. I never opened the code myself
@LizardBuild makes the snapshot → fork path cheap enough that throwing agents at the same bug stops being a waste of money
Pro is $19/mo with $19 credits inside, first week free
Link: lizard.build/snapshot
Instead of debugging it myself I took a snapshot of one Lizard sandbox, forked it into 50 copies in under a second each, and let every Claude Code try a different fix
The first one that got the tests green won
The and cost nothing
Total runtime on the winning path was a few minutes. I never opened the code myself
@LizardBuild makes the snapshot → fork path cheap enough that throwing agents at the same bug stops being a waste of money
Pro is $19/mo with $19 credits inside, first week free
Link: lizard.build/snapshot
I pointed 100 sandboxes on @LizardBuild at the same messy repo and told each Claude Code to find and fix one class of bugs
One looked for missing error handling, another for flaky tests, another for dead code, another for unsafe SQL. They all started at the same time
Median batch was $1.62/h while they ran. As soon as an agent finished its pass and opened a clean PR, I paused that sandbox and it dropped to $0
By the end of the hour I had 73 PRs sitting ready, each with a different fix and a short note on what it changed. I didn’t babysit a single process
Pro is $19/mo with $19 credits inside. First week free with $5 to try
Link: lizard.build/sandboxes/code
10
2
30
913
Lyren_o retweeted
I pointed 100 sandboxes on @LizardBuild at the same messy repo and told each Claude Code to find and fix one class of bugs
One looked for missing error handling, another for flaky tests, another for dead code, another for unsafe SQL. They all started at the same time
Median batch was $1.62/h while they ran. As soon as an agent finished its pass and opened a clean PR, I paused that sandbox and it dropped to $0
By the end of the hour I had 73 PRs sitting ready, each with a different fix and a short note on what it changed. I didn’t babysit a single process
Pro is $19/mo with $19 credits inside. First week free with $5 to try
Link: lizard.build/sandboxes/code
One looked for missing error handling, another for flaky tests, another for dead code, another for unsafe SQL. They all started at the same time
Median batch was $1.62/h while they ran. As soon as an agent finished its pass and opened a clean PR, I paused that sandbox and it dropped to $0
By the end of the hour I had 73 PRs sitting ready, each with a different fix and a short note on what it changed. I didn’t babysit a single process
Pro is $19/mo with $19 credits inside. First week free with $5 to try
Link: lizard.build/sandboxes/code
You shipped an update. 1,000 users lost their files. Not on Lizard
Every app that gives users their own agent hits this sooner or later: you push a new version, rebuild the environments, and whatever your users made inside them is gone
On Lizard each user gets their own s live on a persistent volume at /workspace, not inside the sandbox. Ship a new version, snapshot it once, and every user's sandbox comes back on it in 1 to 3 seconds with the same volume attached
lizard.build/sandboxes
> the sandbox is new, the user's files stay exactly where they were
> no migration scripts, no copying data between machines
> each user's sandbox is its own isolated microVM
> billed per second while it runs, and a paused sandbox costs nothing
Ship as often as you want. Your users won't notice anything except the new features
19
1
45
2,070
I pointed 100 sandboxes on @LizardBuild at the same messy repo and told each Claude Code to find and fix one class of bugs
One looked for missing error handling, another for flaky tests, another for dead code, another for unsafe SQL. They all started at the same time
Median batch was $1.62/h while they ran. As soon as an agent finished its pass and opened a clean PR, I paused that sandbox and it dropped to $0
By the end of the hour I had 73 PRs sitting ready, each with a different fix and a short note on what it changed. I didn’t babysit a single process
Pro is $19/mo with $19 credits inside. First week free with $5 to try
Link: lizard.build/sandboxes/code
One looked for missing error handling, another for flaky tests, another for dead code, another for unsafe SQL. They all started at the same time
Median batch was $1.62/h while they ran. As soon as an agent finished its pass and opened a clean PR, I paused that sandbox and it dropped to $0
By the end of the hour I had 73 PRs sitting ready, each with a different fix and a short note on what it changed. I didn’t babysit a single process
Pro is $19/mo with $19 credits inside. First week free with $5 to try
Link: lizard.build/sandboxes/code
You shipped an update. 1,000 users lost their files. Not on Lizard
Every app that gives users their own agent hits this sooner or later: you push a new version, rebuild the environments, and whatever your users made inside them is gone
On Lizard each user gets their own s live on a persistent volume at /workspace, not inside the sandbox. Ship a new version, snapshot it once, and every user's sandbox comes back on it in 1 to 3 seconds with the same volume attached
lizard.build/sandboxes
> the sandbox is new, the user's files stay exactly where they were
> no migration scripts, no copying data between machines
> each user's sandbox is its own isolated microVM
> billed per second while it runs, and a paused sandbox costs nothing
Ship as often as you want. Your users won't notice anything except the new features
19
1
45
2,070
Lyren_o retweeted
A simple second brain with folders + clear rules for the agent works better than most complex setups.
Basic vault structure:
- Inbox: raw input
- Sources: processed materials
- Projects: active work
- Memory: key decisions and context
- Rules: one file (BRAIN.md)
The agent d works with the vault. Drop new material into Inbox, it structures it and adds links.
The video shows the minimal setup and example rules. Takes 15-20 minutes and removes the need to re-explain context every time.
Basic vault structure:
- Inbox: raw input
- Sources: processed materials
- Projects: active work
- Memory: key decisions and context
- Rules: one file (BRAIN.md)
The agent d works with the vault. Drop new material into Inbox, it structures it and adds links.
The video shows the minimal setup and example rules. Takes 15-20 minutes and removes the need to re-explain context every time.
16
2
39
1,088