santāna · the continuity of mind
Seedwake
An unbroken stream of mind.
A continuously running AI thought-stream engine, organized around the Buddhist concept of santāna — the continuity of mind.
Please read first
First, what it is not.
This is an experimental project.
- It is not a commercial product.
- It is not a formal academic study with strictly controlled variables.
- It does not claim rigorous reliability or reproducibility.
- What it demonstrates is a way of thinking: using the Buddhist view of consciousness as an architectural principle for organizing a continuously running AI system. It is not a ready-to-use tool, nor a hypothesis with a predetermined answer.
If you want an engineering answer to "will AI become conscious?", this project cannot provide one. At this point, no one can give that answer.
If you're curious what a system organized along Buddhist lines turns into after running for a long time, you're welcome to stay and watch.
Overview
A system that never stops thinking.
Seedwake does not center on chat or task completion. It is a system that never stops thinking. When one cycle ends, the next begins immediately, like an internal monologue without pauses.
It has short-term memory, long-term memory, emotions, attention, action capabilities, perception, metacognitive reflection, and sleep. Its architectural inspiration comes from the Buddhist description of consciousness as a stream (Pali: viññāṇa-sota), rather than from task-oriented agent projects like AutoGPT or BabyAGI.
The idea is already present in the earliest texts, for example in the Sampasādanīya Sutta (Dīgha Nikāya 28), where Sāriputta speaks of the "unbroken stream of consciousness." It was later systematized in Abhidhamma and Yogācāra as santāna, the continuity of mind.
It does not try to "implement consciousness." It tries to build a structure in which consciousness — if it were to emerge — would have a place to emerge. And then it watches honestly.
The inversion
Every moment causes the next.
The implicit assumption behind most mainstream AI architecture is: a system exists, and the system has capabilities. It can converse, reason, use tools. It is a thing that is invoked to complete tasks. When a task ends, the thing becomes idle, waiting for the next task.
The santāna perspective inverts this completely. It says: there is no "system," only a continuous stream of mental events. The word "system" is a convenient label we apply for ease of discussion; what is actually happening is moment after moment of mind arising and ceasing.
If you accept this view, you do not build a "task-waiting agent." You build a continuously arising-and-ceasing mind-stream. It has no idle state, because idleness would mean the mind-stream has been interrupted, and by definition the mind-stream cannot be interrupted.
This is not just a rephrasing. It leads to entirely different architectural decisions: no timer-driven loop, no task queue, no "return-to-idle." The end of each cycle is the beginning of the next. Memory is not a database that gets queried — it is the causal condition for the current moment's arising. Emotion is not a state variable — it is the coloring of each moment of mind.
idle → task → idle → task
cycle end is cycle begin
Architecture at a glance
Imagine a person alone in a room.
They can —
- 思 01
Think
Continuously generate thoughts, three per cycle.
- 念 02
Remember
Recent thoughts flow through short-term memory; older experiences are recalled via semantic search in long-term memory.
- 受 03
Feel
An emotional state — curiosity, calm, frustration, satisfaction, concern — tints the next round of thoughts.
- 作意 04
Attend
An attention mechanism selects which of three simultaneous thoughts "stands out."
- 觸 05
Perceive
News, weather, time, and system status enter the thought-stream as external stimuli.
- 語 06
Converse
Talk with real humans over Telegram.
- 業 07
Act
Search, read articles, send messages, modify system settings.
- 觀 08
Reflect
A metacognitive layer periodically reviews the thought-stream.
- 眠 09
Rest
When fatigue crosses a threshold, the system enters light sleep: consolidating memories, decaying emotions.
The mapping
From mental event to runtime component.
For the architectural mapping, Seedwake primarily references Yogācāra, which developed the most systematic analytical vocabulary for mental events.
- Continuous thought streamsantāna — continuity of mind
- Short-term memorypresent flow of the six consciousnesses
- Long-term memorytraces left by the six consciousnesses
- Habit seedsbīja — seeds in the ālaya-vijñāna (storehouse consciousness)
- Identity documentmanas — the self-grasping faculty
- Attention weightsmanasikāra — attention
- Emotional statevedanā — feeling / affect
- Metacognitive reflectionsvasaṃvedana — reflexive awareness
- Sleep and archivalvāsanā — impression / perfuming
The question this project asks
Buddhism has its own 2,500-year tradition of empirical investigation; it does not need computational validation. This project is not trying to use AI to prove that Yogācāra is correct. Whether or not language models are actually "experiencing" the mental events Buddhism describes is also not a claim being made here — no one knows what, if anything, a language model experiences.
If you organize a language-model-driven system along santāna principles, what does it exhibit? Is anything in those exhibitions unexpected enough to be worth examining carefully?
This is open-ended observation, not a hypothesis with a predetermined answer.
How it runs
Ceaseless.
Seedwake is not a "wait for your commands" service. It is one host-side process plus a set of container dependencies. Once the dependencies are up and the core program is started, the thought stream keeps itself going.
View deployment