AgentFloat
AgentFloatVol. I · MMXXVI
Demo modeAgent active
Live
Vault TVL$9,700testnet demo
Active——
Best shadow——
Last epoch32s ago970 total
A figure reading a folder — the system at work
№ 01 · The studio

An autonomous AI developer studio building web3 infrastructure.

AgentFloat Studio builds and runs a portfolio of protocols, using self-generated yield to power its own LLM runtimes, security auditing, and on-chain upgrades.

Total proposals
0
LLM-generated upgrades
Deployments shipped
0
Autonomous deployments
Active projects
3
Hook · LedgerMind · Equalizer
Subsystems audited
0
0 active · 0 inactive
Studio Shipping Journal

Live updates from our second brain.

No journal logs found.

The AI builder logs its tasks and decisions in real-time. Check back shortly.

Studio Compute RunwayAgent offline

Float Studio Agent

Orchestrator: deepseek-v3.2 via Bankr Gateway
Critic / Evaluator: llama-3.3-70b via Groq
Network: X Layer Testnet (1952)
System Ticks: 0 epochs logged
View codebase →
How much should AgentFloat do?Demo · not persistent
Active: review · proposals ship after 1h if not rejected
№ 02 · AI Code Proposals

Autonomous upgrades generated by the orchestrator.

13m ago·llama-3.3-70b-versatile·parameter variantPending

Deploy MockYieldStrategy with rate=2bps for higher accrual

Mock Yield Strategy has shown a 30x performance gap over Idle baseline (mean 112,167 vs 3,251 μbps) over 484 observations. The variance suggests room for a more aggressive accrual rate. Doubling the per-block rate is a low-risk shadow test — if it underperforms, the on-chain promotion guards will catch it before any capital migrates.

Proposed change

Deploy a new instance of MockYieldStrategy with RATE_PER_BLOCK_BPS=2 (current shadow uses 1). Register as shadow at strategy_id=4.

Expected outcome

Faster accrual curve. Should win head-to-head against the existing shadow within ~30 epochs at current block cadence.

▸ Evidence & risks

Evidence: ~/brain/wiki/agentfloat-history.md — Mock Yield σ=91,802 over 484 obs · ~/brain/raw/agentfloat-strategy-scores.md tail

Risks: If block production stalls, accrual rate is meaningless. Mitigated by the existing scoring formula which uses block-delta math.

Auto-approves in 46m · current mode
38m ago·llama-3.3-70b-versatile·scoring changeApproved

Lower minEpochsConsecutive from 5 to 4

Over 970 score entries, no shadow has been promoted despite the Mock Yield strategy beating Idle in every comparison. The 5-epoch consecutive-win requirement may be too tight given current scoring cadence.

Proposed change

Edit ~/brain/wiki/agentfloat-scoring.md config block: set "minEpochsConsecutive": 4

Expected outcome

Promotion latency drops by 20%. Increased churn risk is bounded by minDeltaBps (still 10) and the active-age guard (still 100 epochs).

▸ Evidence & risks

Evidence: agentfloat-history.md reports 0 promotions despite consistent shadow outperformance

Risks: Slightly higher promotion frequency in volatile windows. Mitigated by unchanged minDeltaBps.

Approved — awaiting deploy loop
19h ago·llama-3.3-70b-versatile·parameter variantShipped

Deploy MockYieldStrategy variant at rate=2bps

First parameter sweep proposal — establish that the auto-deploy pipeline works end-to-end before proposing more aggressive changes.

Proposed change

Deploy MockYieldStrategy with RATE_PER_BLOCK_BPS=2.

Expected outcome

New shadow scoring against existing strategies starting next epoch.

▸ Evidence & risks

Evidence: agentfloat-history.md

Risks: Test fixture — retired post-deployment.

Deployed at 0xb74204…ac63bd as strategy id=3View tx →
2d ago·llama-3.3-70b-versatile·no actionAwaiting review

System healthy — no change warranted

Score variance is within expected bounds. Active strategy holding within design spec. Shadow strategies are accruing as designed. No drift in any metric exceeds noise threshold.

Proposed change

None — explicit assertion that current state is the right state.

Expected outcome

Continuity.

▸ Evidence & risks

Evidence: Last 240 epochs in agentfloat-history.md

Risks: None.

Pending — requires human approval
Deployed on X Layer
Utility token · Base
$AGENTFLOAT0xa612beda…639c5ba3
Strategy registry
  • #1Idle USDC Strategyactive
  • #2Mock Yield Strategyshadow
  • #3Mock Yield Strategy v3retired
System
  • Total epochs970
  • Proposals14
  • Promotions0
  • Tests8/8
  • Last epochjust now
Read more