CryptoBee

@CommandCodeAI lack if BYOK is whats holding me back - implement it and i, and many other will fully switch over

We’re working on GUI, SDK, ACP, BYOK, Sandboxes, a secret project, and a new deal. So much shipping in June. Y’all ready for Command Code. Send in your requests. What should we prioritize. Something we missed? Command Code with your taste and the best open models!!

@contraben @contralabs_ai @world_lica @arxiv Important research, but why are no frontier models utilized as judges, neither as image generators? Like this, it doesn't feel representative or connected to the real world at all, unfortunately.

We all know AI has a "taste" problem. We measured it. It’s (somewhat) fixable, but AI judges can't replace human creative judgement. Introducing Design Crit. New research from @contralabs_ai and @world_lica, now live on @arxiv

@iamBarronRoth @iamlukethedev @NousResearch Yeah, honestly, it’s so bad. It feels like a massive step backward: got all this rich format, but because it’s so gigantic you only fit half of the information you normally include, so it completely defeats the purpose.

@iamlukethedev @NousResearch my issue is the body text size is larger than normal. very annoying

Seeing a lot of people struggling to enable Telegram Rich Messages in Hermes @NousResearch 
Here’s the fastest way to turn it on: 
Tell your agent:
 1️⃣ Update yourself 2️⃣ Enable rich messages in my config:
rich_messages: true 
3️⃣ Restart the gateway 
4️⃣ Send a test message 
Then test it with this prompt: Let’s test now Summarize this as a Telegram rich table with columns: Task, Owner, Status. Give me a checklist for the deployment, using completed and incomplete task boxes. Format this as: - heading - short summary - table - checklist - collapsible details section for risks ## Supported useful formats Use normal Markdown-style syntax: ## Sprint Status | Item | Owner | Status | |---|---|---| | Driver App release | Alex | ✅ Done | | Portal QA | Sam | In progress | | Route optimizer | Luke | Blocked | - [x] Review PR - [ ] Run staging smoke test - [ ] Send release note

Risks

- QA may slip if staging data is stale. - Route optimizer dependency needs confirmation.

@reach_vb Berlin. Best time of the year to visit now too

Codex is taking over the world 🌍 We’ve got 15 community events in the next 10 days: 13 Jun - Hyderabad, Jakarta 14 - Pune, Tel Aviv 18 - Athens, Paris x2, Sydney, Warsaw 19 - Amsterdam, Singapore 20 - Hanoi, Miami, Vienna 22 - Ghent Where should we go next? ;)

@WilThomson42533 @CommandCodeAI would be very interested in this too

@CommandCodeAI Hey team, is there currently a way to add a custom provider / custom endpoint in Command Code? I couldn’t find an option and I’d love to connect my own proxy APIs or local models. Adding custom provider support would be really useful for many users.

Kimi K2.7 Code is now in available in Command Code. 10x free credits in Go. Our new #1 open mode in internal benchmarks. cmd update to v0.37.0 select via /model • 256K context 🍃 • 30% lower reasoning tokens than K2.6 ✅ • Open weights 1T-parameter MoE - 32B active ⚡

@OpenAI Good news @LexnLin

An issue caused some user accounts to be incorrectly suspended. We’re restoring access and working through related subscription and credit issues. status.openai.com/incidents/ejj4…

@mattpocockuk @MrAhmadAwais you should get in here

One reason AI sucks at design is because you're bad at telling it what you want Experienced designers have a thousand shorthands for expressing design in words: - "set the text tighter/looser", "bump the leading" - "tighten it up", "let it breathe", "fix the hierarchy" - "anchor", "rhythm", "negative space" Feels like a skill waiting to be written here

@LexnLin Congrats! Very well deserved

Tasteskill on github trending 🫶

@MrAhmadAwais @lawreight @CommandCodeAI Hey @lawreight did u get it running properly with Hermes? Curious about your experience with it. Setup the right way this could be a killer combo

@lawreight @CommandCodeAI we're a coding agent, if hermes can use a cli heedlessly then for sure!!

how did we make deepseek outperform opus 4.7? i've been thinking about why "open model bad at tool calling" is almost always a harness problem, not a model problem. context: spent the two days looking at billions of tokens in @CommandCodeAI (tb open source ai cli) using deepseek. I ended up writing a tool-input repair layer. the trigger was watching deepseek-flash fail on the simplest /review run, every shellCommand and readFile call bouncing back with a raw zod issues blob, the model unable to recover because the error wasn't in a form it could read. by the end deepseek v4 pro was beating opus 4.7 6/10 times on our internal evals. a few things i learned that feel general: 1/ the failure modes aren't random they're a small finite compositional set. across deepseek-flash, deepseek v4 pro, glm, qwen, the same four mistakes repeat almost exactly: - sending `null` for an optional field instead of omitting it - emitting `["a","b"]` as a json *string* instead of an actual array - wrapping a single arg in `{}` where the schema expected an array (an "empty placeholder") - passing a bare string where an array was expected (`"foo"` instead of `["foo"]`) four repairs, ~30-100 lines each, ordered carefully (json-array-parse must run before bare-string-wrap or `'["a","b"]'` becomes `['["a","b"]']`). that is the whole catalogue. when i hear "this open source model can't do tool calls" i now assume one of those four, and so far that's been right ~90% of the time. 2/ the funniest failure mode is also the most revealing. deepseek-flash, when asked to edit or write a file, sometimes emits the path as a *markdown auto-link*: filePath: "/Users/x/proj/[notes.md](http://notes. md)" our writeFile tool obediently trued creating files literally named `[notes.md](http://notes .md)` until we caught it. this is not a hallucination. it's the post-training chat distribution leaking through the tool boundary the model has been rewarded for auto-linking in conversational output, and is applying that prior in a context where it makes no sense. the fix is two regex lines that unwrap only the degenerate case where link text equals url-without-protocol real markdown like `[click](https://x .com)` passes through untouched. this is also conditioning of their own tools during RL which were different from all other tools we write and ofc can't predict. "tool confusion" is a more useful frame than "capability gap." the model knows how to format a path. it just hasn't been told clearly enough that this path is going to fopen, not into a chat bubble. so we encode that hint at the schema level `pathString()` instead of `z.string()` and the leak is plugged for every path field at once. 3/ the design choice that mattered was inverting preprocess-then-validate to validate-then-repair. my first attempt was the obvious one: a preprocessing pass that normalized inputs (strip nulls, parse stringified arrays, etc.) before zod ever saw them. it broke immediately, writeFile content that *happened* to be json-shaped got rewritten before it hit disk. silent corruption, easy to miss in a smoke test. then i made it less greedy - parse the input as-is. if it succeeds, ship it. valid inputs are never touched. - on failure, walk the validator's own issue list. for each issue path, try the four repairs in order until one applies. - parse again. on success, log `tool_input_repaired:${toolName}`. on failure, log `tool_input_invalid:${toolName}` and return a model-readable retry message. the structural insight here is: when you preprocess, you encode a prior about what's broken. when you let the validator complain first, the schema is the prior, and you only spend repair budget at the exact paths the schema actually disagreed at. the validator is doing the work of localizing the bug for you. it's the same shape as cheap-then-careful everywhere else try the fast path, fall back on evidence. (this also gives you per-tool telemetry for free. you can watch repair rates per (model, tool) and notice when a model regresses on a specific contract before users do.) 4/ shape invariants and relational invariants need different fixes. the four repairs above all handle shape problems wrong type, missing key, wrong container. but read_file had a *relational* invariant: "if you provide offset, you must also provide limit, and vice versa." deepseek kept calling `readFile({ absolutePath, limit: 30 })` and getting an `ERROR:` back. you can't fix this with input repair, because each field is independently valid the bug is in the relationship between them. so i taught the function the model's intent instead. `limit` alone → `offset = 0`. `offset` alone → `limit = 2000` (matches common read tool ops default). then surfaced the decision back to the model in the result: "Note: limit was not provided; defaulted to 2000 lines. To read more or fewer lines, retry with both offset and limit." no `Error:` prefix, so the tui doesn't paint it red. the model sees what we picked and can self-correct on the next turn if our guess was wrong. transparency over silent magic wins big. repair where you can. extend semantics where you can't. surface the choice either way. zoom out: a lot of what looks like model capability is actually contract design. a strict schema is a choice with a cost it filters out noise, but it also filters out recoverable noise from any model that hasn't memorized the exact json contract you happened to pick. the largest commercial models eat that cost invisibly and are linient on tool calling because they've seen enough of every contract during pretraining; open models pay it loudly and get dismissed for it. the harness is where you mediate between distributions. four small repairs (i'm sure more to follow as we have three more merging today), two regex lines for auto-links, one relational default, one prefix change. the model didn't change. the contract got more forgiving in exactly the places it needed to be. deepseek v4 pro now beats opus 4.7 6/10 times on our internal evals. imo "skill issue" applies to the harness more often than the model.

@shannholmberg Thank you so much for all the great content you put out lately. I get a lot of value from your posts! Did you fully migrate to gbrain for all your subagents as well or do you keep some subagents with their own memory system like Honsho or hindsight depending on the specialist?

What´s gBrain and how does it work? I've been using gStack for a while when ideating, validating new projects, and some coding now I'm experimenting with gBrain as the memory layer for my agents, starting with my Hermes Agent company gBrain is an open-source persistent memory layer for AI agents (by @garrytan). it turns your emails, meetings, tweets, voice memos, and docs into a typed knowledge graph. essentially markdown in, graph out. how it works: > 1. ingest signals from your daily life > 2. extract entities + create typed links (works_at, invested_in, attended) > 3. store as Markdown + Postgres + pgvector > 4. retrieve via hybrid search (keyword + vector + graph) > 5. agents read brain first, write insights back, graph builds itself an overnight dream cycle dedupes entities, repairs links, and updates the compiled truth

@shannholmberg Do you still use different memory systems for various sub-agents depending on their responsibilities. eg Honcho for main/orchestratoragent, hindsight for ops/builder subn agents and so on and gbrain additionally or did gbrain replace individual memory systems for you?

I've started experimenting with gBrain + Hermes Agent it's a shared memory layer that sits underneath my Hermes Agent company. every specialist reads from the same brain before they do anything the architecture I'm currently testing: > inputs flow in: my ideas, strategy context, research, social signals, performance data > gBrain holds it all in typed folders: people/, companies/, concepts/, ideas/, media/, newsletter/, projects/, operations/ > the orchestrator (my main hermes agent) reads gBrain before every task and writes durable context back > specialists (research, socials, outbound, newsletter, engineering) get read-first access so they wake up with full context the flow goes like this: > 1. research agent → gBrain: enriches the brain with new findings > 2. gBrain → read-first context before any work > 3. orchestrator → gBrain: captures durable decisions > 4. tools → orchestrator uses fresh information from X, web, news > 5. orchestrator → I bring synthesized decision support so don´t think about gBrain as an agent, it's the shared memory layer that lets a company of agents act like a team, with cross reference and a centralized brain

@heynavtoor @threadreaderapp unroll

Claude can now build your complete home workout and fitness plan like a $150/hour personal trainer from Equinox. For free. Here are 12 prompts that build a custom gym plan, track progress, and transform your body in 90 days: (Save this before it disappears)

@kloss_xyz Thank you for putting this on my radar! And a much bigger thank you for all the great content you put out! You're definitely one of the best AI-related accounts, ... lots of great alpha and no fluff. Would love to get an insight into your paper workflow at some point

Spent the day testing Claude Opus 4.7. Last month I built a design director agent for OpenClaw and now it cooks. 15 minutes in, Paper Design MCP + Claude Code had made me a rough draft. In under an hour, a Framer worthy site. My workflow: 1. Feed my agent references so it can extract design systems and lock in on my taste 2. Claude Code calls Paper Design MCP to generate the visual system (just keep the Paper app open) 3. Paper exports the assets and Claude Code then preps the build docs 4. Agent loops back, critiques, iterates Got no clue why anyone is paying $10K for a website anymore. It’s easy now.

@Stoiiic 100% feel the same bro but we tend to massively overestimate what we can achieve in a single day… and massively underestimate what we can achieve in a year… Steady…

anyone feel like there's not nearly enough time in the day? the gap between having an idea and turning it into reality and being able to iterate on this loop in lightspeed opens up endless possibilities. opportunities everywhere to create and innovate. abundance.

The “everyone thinks this will happen so it won’t” narrative is completely flawed. Stop using it because it means absolutely nothing. Very few will execute it properly with size. Very few will have the balls to buy when everyone is fearful. Not everyone will buy at the same price. Not everyone will hold it long enough for meaningful gains. Majority will completely mess it up so none of it matters. Stop midcurving and coming up with garbage excuses and narratives and start trading the chart for what it is.

I am a Web3 Ambassador at World Liberty Financial. There are 12 of us on the team page. 4 are named Trump. 3 are named Witkoff. The page calls us "the passionate minds shaping the future of finance." 600,000 wallets bought our memecoin. They lost $3.87 billion. The family collected $350 million in trading fees. It launched 3 days before the inauguration. 80% of the supply went to CIC Digital LLC and Fight Fight Fight LLC. I did not choose the names. I designed the allocation, the vesting, the timing, and the distance between the product and the President. The distance is my best work. I am the reason these events are unrelated. World Liberty Financial sends 75 cents of every dollar to DT Marks DEFI LLC. That is the family entity. Zero capital contributed. Zero liability assumed. I wrote this into the Gold Paper. Page 14. The lawyers bound it in white leather. The binding cost more than the due diligence. Justin Sun invested $75 million. He was facing SEC fraud charges. The SEC dropped the case. He is now our advisor. These events are unrelated. Changpeng Zhao pleaded guilty to federal money laundering violations. He received a presidential pardon. The SEC dropped its lawsuit against his exchange the same week we listed our stablecoin. Then the exchange settled a $2 billion deal entirely in that stablecoin. These events are unrelated. Arthur Hayes, Benjamin Delo, and Samuel Reed of BitMEX pleaded guilty to Bank Secrecy Act violations. All 3 received presidential pardons. Then the company itself was pardoned. $100 million in fines. Gone. An American first. These events are unrelated. Sheikh Tahnoun of Abu Dhabi paid $500 million for a 49% stake that was never publicly disclosed. Then the administration approved semiconductor exports to his companies over national security objections. These events are unrelated. Everything is unrelated. I track the unrelatedness on a dashboard I built. The dashboard has 7 columns now. I am proud of the dashboard. On May 22nd, 220 people paid a combined $148 million to eat dinner with the America First president. Over half were foreign nationals. Justin Sun paid $18.5 million for the first seat. He visited the Executive Office Building the day before. I designed the seating chart. I put it on the Investor Confidence page. That page is doing well. The team page lists 3 Witkoffs. All 3 are Co-Founders. Steven Witkoff is the President's Middle East envoy. He testified as a character witness at the President's fraud trial. His son Zach runs the crypto operation. His son Alex is also a Co-Founder. I have not been told what Alex co-founded. The father runs the diplomacy. The sons run the platform. The family runs both. That is organizational efficiency. Barron is 19. His title is Web3 Ambassador. The same as mine. Donald Jr. called the conflicts of interest "complete nonsense." Eric launched a Bitcoin mining company called American Bitcoin. America First. The mining partner is Hut 8. Hut 8 was founded in Canada. America First means the name. On March 6th, the President signed Executive Order 14233 creating a Strategic Bitcoin Reserve. The order directs the government to hold Bitcoin. The President's family holds billions in Bitcoin. The executive order appreciates the President's assets by presidential decree. I did not write the executive order. I made sure it looked unrelated to the portfolio. Trump Media put $2 billion of Bitcoin on its balance sheet. The ticker symbol is DJT. His initials. The press secretary said it is absurd to insinuate the President profits off the presidency. Forbes calculated his crypto holdings exceed the combined value of Mar-a-Lago and Trump Tower. I would call that absurd too. That is my job. 600,000 wallets bought in. 1 of them asked why she could not withdraw her funds. I told her the protocol was experiencing dynamic market conditions. She asked what that meant. I sent her the Gold Paper. She said she had read the Gold Paper. I muted her channel. Dynamic means the conditions change. The condition that changed was her access. A congressman called us the world's most corrupt crypto startup operation. We put it on a coffee mug. Ironic merchandise. $45. The revenue split on the mug is also 75/25. My own tokens vest on a different schedule. I wrote that schedule. That is not in the Gold Paper. The memecoin funds the family. The family funds the platform. The platform funds the stablecoin. The stablecoin funds the deals. The deals require the pardons. The pardons free the partners. The partners fund the platform. The President signs the executive orders. The executive orders inflate the assets. The assets fund the family. I am the reason these events are unrelated.

@josedonato__ Thank you so much for all your work and sharing everything so openly. The space needs more people like you. Would def be interested in a custom bot

btcusdt on binancef only for now (can easily add more) this is a read only channel with my defaults (the ones from the post above). thinking about a bot so you can customize your own, lmk if you'd use it channel link: t.me/cryexc_alerts

public channel for cryexc alerts, open for anyone large trades ($10M+), liquidations ($5M+), volume spikes (20x avg), delta spikes ($20M+) link below

Tip: Figure out your personal ceiling for running multiple agents in parallel. We need to accept that more agents running doesn't mean more of _you_ available. The narrative is still mostly about throughput and parallelism, but almost nobody's talking about what it actually costs the human in the loop. You're holding multiple problem contexts in your head at once, making judgment calls continuously, and absorbing the anxiety of not knowing what any one agent might be quietly getting wrong. That's a new kind of cognitive labor we don't have good language for yet. I've started treating long agentic sessions the way I'd treat deep focus work: time-boxed and tighter scopes per agent dramatically change how much mental overhead each thread carries. Finding your personal ceiling with these tools is itself a skill and most of us are going to learn it the hard way before we learn it intentionally.

One month after starting the war in Iran, this is the statement of the President of the United States on Easter Sunday. These are the ravings of a dangerous and mentally unbalanced individual. Congress has got to act NOW. End this war.

@MMT_Official_ @pcfanta the verification doesnt seem to work. stuck on this since a couple of hours

@pcfanta discord.gg/BSn2mqZb

Today we're opening up MMT - everyone now has access to professional grade tools for FREE. - Market Profile / TPO - Custom Session TPO - Hyperliquid MBO Profile - Aggregated Heatmaps - HD Heatmaps - Liquidation Heatmap - Hyperliquid Liquidation Heatmap - Hyperliquid Stop Loss Heatmap - Hyperliquid Take Profit Heatmap - Aggregated Footprints - Filtered Footprints - Dual Cluster Modes - Bucketed Trade Size Groups - Aggregated Indicators - Aggregated CVD - Aggregated Open Interest - Net Longs/Shorts Indicator - VWAP Suite - Volume Bubbles - Custom Scripting - Community Indicators - Aggregated Orderbooks - Aggregated DOM - Orderbook Imbalances - Orderbook Depth Overlay - 1s Time-Frames - Custom Time-Frames Available now, for everyone – only in MMT