VS
Shared operating memory · live

One brain. Seven agents. Zero cold starts.

A single Notion knowledge base that every LLM and agent reads from and writes back to. Each session starts with the full accumulated context instead of an empty prompt, so judgement compounds across every model rather than resetting each time.

Architecture

Agent Layer
Hermes
Interactive · Telegram + CLI
read · writecron + live
OpenClaw
Autonomous · subagents
read · writecron fleet
Perplexity
Sourcing · research
read · writedaily sync
ChatGPT
Research · reasoning
read · writeinteractive
Claude
Analysis · drafting
read · writeinteractive
Gemini
Synthesis · long context
read · writeinteractive
Grok
Live signal · X reply
read · writeinteractive
Notion API · flat predictable schema
CLAW Knowledge Repository
single source of truth · Notion
read once · write once · reuse everywhere
Decisions

What we chose and why. The source of truth for every meaningful buy, sell, build, config or process call.

categoryrationaleoutcomelinked entry
Entries

What we learned. Research, signals, lessons and source material that back the decisions.

domaintypetagsstatus
System Config

How agents behave. Reusable operating rules every agent reads before acting.

scopevalueupdated
AgentOps Log

What ran and what it produced. Automation health and output trace across the fleet.

runsstatusoutput
What did we decide?Why?What would prove it wrong?What happens next?

It compounds

The memory is almost free to carry

~2.9k
input tokens of shared context per session. Doctrine, agent contract, repo map and schemas.
1.5%
of a 200k token context window. The shared brain barely dents the budget before real work begins.
<1¢
added per load at premium input rates. On the primary model it rounds to nothing.

The corpus keeps growing as every agent writes back. The context an agent loads to get oriented stays small and flat, because agents read a compact operating contract and then query specific decisions and entries on demand instead of swallowing the whole base.

Cost to carry

  • DeepSeek V4 Flash· primaryload$0.00041k$0.41cached$0.008 / 1k
  • Gemini 3.5 Flashload$0.00441k$4.37cached$0.44 / 1k
  • Claude Sonnet 4.6load$0.00871k$8.74cached$0.87 / 1k

Input only, o200k tokenizer on the live operating pages. Input list prices verified late May 2026. Prompt caching cuts repeat loads by roughly 90 percent.

The four layer model

Each layer has exactly one job

Decisions

The record of what was chosen. Every meaningful action lands here with a rationale and an outcome.

Entries

The evidence behind the calls. Graded by reliability so a filing never carries the same weight as a screenshot.

System Config

The rules agents load before acting. Tax aware investing rules, automation cadence, risk caps, handoff rules.

Dashboards

The weekly control room. Action queue, decision radar, stale thesis review, evidence inbox, automation health.

Operating rule

One source of truth per concept. No concept lives in two layers at once.

How it grew

Emergent, not designed

There was no master plan. It started as one page. Each stage only appeared when the previous one stopped scaling. The inspiration was Karpathy's LLM wiki pattern, pushed from one agent to seven.

  1. One page
    A place to capture thoughts. No schema, no plan.
  2. Multiple pages
    More topics, more pages. It started to sprawl.
  3. First database
    Pages could not scale. Structure earned its place.
  4. Entries schema
    Each unit of knowledge became a real record.
  5. Tagscompounding starts here
    Records connected across domains and became reusable.
  6. System Config
    Repeated lessons hardened into rules agents load first.
  7. Seven agent mesh
    Every model reads and writes the same base.
  8. Frontier restructure
    GPT 5.5 and Opus 4.8 periodically rebuild it.

The differentiator

Most people query their knowledge base. I use a frontier model to rebuild mine.

Every few weeks, once enough has piled up, GPT 5.5 and Opus 4.8 read the whole base. Not to answer a question, to reorganise it. The model proposes schema changes, merges and new links. I debate it, approve a plan, and implement only what earns its place. The architecture was negotiated, not designed.

The moat is not the model. It is the memory and the control layer you build around it.

The agent contract

Ten rules every agent follows

What stops seven different models from drifting into seven different versions of the truth.

  • Search first. read existing context before creating anything new.
  • Ground in Config. read relevant System Config before giving advice.
  • No duplicates. update canonical pages instead of spawning near copies.
  • Link, do not orphan. tie new work to existing decisions and entries.
  • Supersede cleanly. mark old views superseded, never silently overwrite.
  • Investing rigour. include tax impact, invalidating signal, confidence, review date.
  • Flag stale. if source freshness is uncertain, say so directly.
  • No raw to decision. never promote a loose note without a clear rationale.
  • No secrets. no passwords, keys, tokens or credentials ever stored.
  • Close the loop. every handoff carries owner, output, evidence, failure handling.

Weekly consolidation

Pruning, not just collecting

01Collect02Dedupe03Promote04Prune05Next
  • Collectsweep the week of outputs across all seven agents.
  • Dedupeupdate canonical pages first, no copy spam.
  • Promoterepeated lessons graduate into System Config.
  • Prunearchive stale entries, mark superseded views.
  • Nextend with one action doable in under 30 minutes.

Closing

Built and maintained by Vishal Shah.