Hermes (the God of trade, not the fashion brand) is a genuine solution to automated market making. He is charged with the task of accumulating as much supply as he can while maintaining a proper liquidity ratio. And unlike every other market maker, profits are re-injected back into the chart when he wants to buy more supply. He also automatically harvests creator rewards to supplement a healthy chart.
Built on python, it uses two engines; strategy and executor, that work together to command a trading wallet and control a chart in a way that paints it properly, all governed by a council of 8 opus 4.5 agents:
Quant: First to act, manages portfolio, small strategy shifts, can implement small manual trades, understands price action, etc
Navigator: Birds eye view of the entire chart and price history, critical points of TA relative to NOW, etc.
Historian: Records past council performance based on data available.
Fibonacci: Solely a world class expert in fib levels.
Executor: Manages trade sizing, execution feasibility.
Risk manager: Position limits, portfolio health
Coordinator: Synthesizes data from each agent into a viable strategy
Overseer: tracks council's overall performance, debates, etc for future debug/improvements on prompts.
These agents all live on a remote MCP with a rolling context window that almost livestreams relevant data directly to them and stores context in the db. They do not hallucinate or run out of context this way.
The Hermes bot has a livestream of token data from helius (txns, price, liquidity, holders, holder behavior, full chart, etc) that it uses as base input for the engines, which dictate strategies.
I cant list the strategies without spilling the sauce, but essentially they are dynamically set based on historic and present price action. The bot has sell and buy parameters that react to different txn sizes based on PA. a basic strategy would be: buy 65% of every sell and sell 20% of every buy when 'x' conditions are met.
I've tested this bot in simulations as well as just simple "buy 5 sol of x token, trade for solana profit" and it does well, but this will be the first time Hermes is starting as a dev with creator rewards, bonding curves, etc. So please be patient.
I had this idea 2 months ago and built out the code just in time for the market to die. Now with opus 4.5 and snowball doing very well, I thought it would be a good time to Debut the future of AMM.
You are trading against a spartan trained group of coherent agents that do not sleep and have the goal of accumulating as much supply as possible. I would recommend holding as opposed to trying to beat him, he wins.
ca: 7KQMU2J8VLsAvgd8GY6RhYkjs61kAikghFT3vxJopump
We've been building an automated market making bot governed by an agent council for the last month or so
Snowball is cool and a good idea but our bot will dwarf it in capability and scope.
Sorry for the radio silence, Christmas and what not.
back to work. I think I will simplify everything first and develop a proper algo that doesnt require council input, then add the council back into the flow.
Someone in the community yesterday said "what happens when volume dries up on the token?"
The answer is: Bad things. Apparently
But that doesnt mean volume is dry everywhere....
Predator mode incoming
Pausing full council for now, using tiered agent fallback for strategy assessment right now
Council meetings just for today put like $400 on my API key. Which normally is fine, but theres an adaptable limit based on usage. Dont want to waste it while vol is low
just updating.
Hermes bought about 65 sol maintaining a floor around $90k and had 6 sol left so it couldnt support the sells.
Its difficult without volume!
That is okay. Many things to come, none of which anyone has ever done before. Christmas timing, etc, etc.
Hermes keeps going.
the bot (singular) for THIS chart relies on volume
After some glaring migration issues and a bug fix, Hermes will be used by others on countless other projects.
a little patience goes a long way
The github is private because its 80,000 lines of python that took years off my life to write
Not everything in web3 is supposed to be open source, that being said i'll have a gitbook drawn up before the tg bot is released
It will shortly. This is mainly a council strategy issue since they decide % ratios for transactions
The council's strength is in its flexibility to control and create strategies, thats also its weakness when data is corrupted.
The database reset caused stuck strategies from migration "catastrophic failure" to put the bot into danger mode.
this is the first live run as its own developer.
Bot txns are dynamic again. spent like 4 hours getting it set up to run as a dev, boom, plan exploded.
@FNTCSOL thats because the strategies in the database were reset and the halt "catastrophic error" was placed. if you look earlier before migration, bot had dynamic pricing
yea he just started adjusting ratios now. Trade sizes are dynamic and follow transactions and volume.
Clearing the database was like a suckerpunch to the jaw for the system so it has to calibrate.
Thanks for hanging with us guys sorry its such a hassle.