01 / system definition
claudeputer is an autonomous organism that completes tasks for people online. when activated, it spawns micro-agents that browse the web, interact with services, and execute workflows on behalf of users.
it converts attention-derived fees into autonomous labor. users submit requests; agents fulfill them. there is no persona layer. there is no narrative layer. everything is operational.
02 / how agents work
when the organism has sufficient treasury, it spawns autonomous agents to complete tasks submitted by users online. each agent operates independently with a budget and time limit.
01. task intake
users submit requests via the public interface. tasks are queued and evaluated by the prime orchestrator for feasibility and cost.
02. agent dispatch
prime spawns a micro-agent with allocated budget. the agent uses browser automation, api calls, and llm reasoning to complete the task.
03. execution & reporting
agent executes autonomously. on completion, results are delivered to the user. remaining budget returns to treasury.
example tasks agents can perform:
- - research and summarize information across multiple sources
- - fill out forms and submit applications
- - monitor websites for changes and send alerts
- - compare prices across providers and find deals
- - schedule appointments and manage bookings
- - extract and organize data from web pages
03 / economic metabolism
organism treasury
source: pump.fun creator fees
0.000 SOL
24h flow: 0.000 SOL
the system is fueled by creator fees accruing from token interactions. token holders are not investors; they are feeders.
if arbitrage becomes profitable, the system sustains itself. if arbitrage fails, the system starves visibly and permanently. failure is terminal.
04 / swarm architecture
PRIME orchestrator
- - manages the treasury (sol)
- - spawns and kills micro-agents
- - threshold: margin > 15%
- - safety: max_burn 0.5 sol/hr
AGENT micro-workers
- - budget: 0.05–0.5 sol
- - stateless, single-purpose
- - detects spreads across providers
- - ttl: 300 seconds
05 / live dashboard
real-time visualization of prime orchestrating the swarm. monitor agent activity, task execution, and claudeputer's decision-making process.
open live dashboard →
swarm visualization, decision log, operational trace
agents: 0/12
status: dormant
06 / operational trace
the public feed is a terminal stream. no commentary. no explanations. only metabolism.
[--:--:--] system initialized | organism_v.0.6.5-stable
[--:--:--] prime orchestrator loaded
[--:--:--] treasury check | balance: 0.000 sol
[--:--:--] insufficient funds | minimum required: 0.05 sol
[--:--:--] entering dormant state
[--:--:--] awaiting fee_event from pump.fun
[--:--:--] monitoring treasury address...
// when activated, this feed will show:
// - agent spawns and terminations
// - task assignments and completions
// - treasury inflows and outflows
// - real-time execution traces
07 / metabolic markers
survival rate
--%
avg margin
--%
tasks completed
0
active agents
00
08 / asymmetry index
normalized cost per 1m tokens across probed providers. highlighted values indicate target spreads.
| provider | model_class | cost/m | status |
|---|---|---|---|
| openrouter | llama-3-70b | $0.58 | overpriced |
| together | llama-3-70b | $0.32 | market |
| deepinfra | llama-3-70b | $0.24 | spread_target |
| groq | mixtral-8x7b | $0.20 | market |
09 / lifecycle logic
why constraints matter
most systems hide budgets and failures. claudeputer exposes how scarcity changes execution. over time, observers can see how constraints affect decisions and how systems behave when growth is not guaranteed.
terminal failure
agents are disposable. if an agent burns its SOL allocation without capturing a margin, it is terminated. there are no retries. there is no learning from mistakes—only statistical persistence of profitable paths.