deliberate decision to make memory task scoped, so not every task knows about every other task. say you initiate a research task on stablecoin yield farming, that task gets better over time through the memory + dreaming loop, then combined later on with agents that can build strategy code or execute transactions
We're building Fabrick as a user-friendly gateway to both the Solana ecosystem and multi-agent intelligence.
v0 (current):
- Sign in with email or passkey, get an embedded wallet provisioned automatically.
- Chat agent with tools for DeFi analysis, token insights, and news.
- Multi-agent fleets for deep research and onchain data analysis.
Targets for v1:
- Dream agent added to fleet. Key memories from fleet runs persist to improve future runs.
- Memory tooling added for agents to query insights about past runs.
- DeFi connectors. Access the full ecosystem of Solana DeFi ecosystem from Fabrick.
- Coding agents to build and maintain strategy code for trading and DeFi strategies.
Find the link to join our Discord server below, or DM @e_ssshadow if you have any feedback or suggestions for what you want to see in Fabrick.
building and more importantly testing in public. the other week I got started on an autoresearch agent for discovering DLMM strategies on @MeteoraAG
I managed to get the agent to produce a strategy based on the volatility capture objective, today im trying out the sharpe ratio objective
Last week, we put out a quick demo of a Solana Mobile app to create, execute, and manage DLMM positions on @MeteoraAG
To showcase the modularity of multi-agent systems on CoralOS and expand the surface of what you can do from the app, we started building a dedicated discovery agent.
It uses the autoresearch loop pattern to run experiments over historical pool data and write a strategy function that's optimised for a defined metric.
In this example: volatility capture.
This standalone agent will eventually be Coralised and swapped with the vanilla strategy agent in the LP app.
Here's a first look at a Meteora LP portfolio management app built for Solana Mobile that coordinates 3 specialised agents over CoralOS.
1. Scout agent - ingests pool data from one or more Meteora pools selected by the user.
2. Strategy agent - takes the user's budget input, receives the pool-level data from the Scout, and stages a set of transaction proposals for the user to select.
3. Operator agent - once the user selects a strategy, it receives the transaction object from the Strategy agent and executes against the Meteora pool/s.
Stay tuned to learn more about our upcoming initiatives to make Seeker the home of agentic applications on Solana.
@depression2019 you just proved the point of the book Misbehaving in one post lmao. if everyone was perfectly rational, then coordinating around 100% is easy on paper, but assuming perfect rationality in every person is irrational so the rational bet becomes 51% is more likely than 100%
Building HyperAgent.
Autonomous trading terminal for @HyperliquidX
Pick a strategy → hit start → it reads the market, executes trades, manages risk, and explains every move.
One terminal. Six strategies. Hyperliquid testnet-safe.
been imessage-pilled the past couple of days
one-shotted (kinda) a simple Kimi K2.6 multi-agent system + @CoralOS_ai agent message bus connector with @budapp
https://t.co/mKZ6KUi4Nk
new session in new tab, you can just have multiple tabs open at once. not big on paralellisation though personally, unless it different sessions working on different branches, ive found that having several parallel sessions writing changes to the same branch leads to some silent trip ups i.e. "agent A silently changed something in page.ts 3s after agent B, agent B carries on not knowing that its change got overwritten and implements something in server.ts that depends on it"
It's good to have this documented as a study, but at the same time this is kinda a "sky is blue" ahh conclusion.
Catering to vibes is a marketing thing, influencers do it to get clicks and sell courses, platforms do it to sell subscriptions and upsell to higher tiers through aspiration capture.
But deep down I think everyone already knew that software development wasn't actually gonna get displaced by hordes of people with 0 technical background and vibes.