Vibe-Trading’s best upgrades for agentic trading
5 Vibe-Trading upgrades show how the agent now covers broker access, swarms, compare tools, caching, and a cleaner CLI.

Vibe-Trading adds broker connectors, swarm tools, caching, comparison, and a cleaner CLI.
Vibe-Trading has grown into a trading agent with real surface area: the repo shows 11.2k stars, 2.2k forks, and a fast-moving release stream. This list covers five updates that matter most if you want to understand what the project can do now.
| Item | What it adds | Scope |
|---|---|---|
| Interactive CLI refresh | Bigger banner, cleaner divider, recap, timing, activity rail | Terminal UX |
| Live swarm status | Inline per-worker state in chat timeline | Multi-agent runs |
| Alpha compare | Ranks named alphas by IC, IR, and hit rate | Research and benchmarking |
| Data cache | Local cache for seven data sources | Backtests and loaders |
| Broker connectors | Direct SDK and MCP broker access | Execution layer |
1. Interactive CLI refresh
Get the latest AI news in your inbox
Weekly picks of model releases, tools, and deep dives — no spam, unsubscribe anytime.
No spam. Unsubscribe at any time.
The newest terminal front door makes the agent feel easier to watch and easier to resume. The 2026-05-23 refresh adds a larger Vibe-Trading banner, a cleaner prompt divider, a recap of the prior turn, post-run timing, and a Claude Code-style activity rail for live work.

That matters because the CLI is still the fastest way to see what the agent is doing before you wire it into a broader workflow. The update is not just cosmetic; it gives you context after each run, which helps when a tool call chain gets long or when you return to a session after a pause.
- Prior-turn recap for quick re-entry
- Run timing shown after the agent finishes
- Activity rail for live progress feedback
2. Live swarm status
When Vibe-Trading launches a multi-agent swarm, the chat timeline now shows each worker's state inline. You can watch waiting, running, done, failed, blocked, and retrying statuses stream in real time, which makes the swarm easier to read without jumping to a separate dashboard.
This is useful for investment committee-style runs where several agents are working at once. The status cards rehydrate from the final result on reconnect or history replay, so you do not lose visibility if the session refreshes mid-run.
- Works for swarms like investment committee, quant desk, and risk committee
- Runtime events are bridged into the session SSE stream
- Preset routing now prefers an explicitly named preset over keyword scoring
3. Alpha compare
The alpha compare feature turns the Alpha Zoo into a head-to-head bench instead of a long catalog. You choose a shortlist of alphas, run them on a universe and period, and get ranked results by IC mean/std, IR, positive IC ratio, or sample count.

The useful part is focus: it compares only the alphas you name, rather than scoring the whole zoo. That keeps the output tighter for research reviews and makes it easier to see which signal is leading and by how much.
vibe-trading alpha compare --sort ir
- Available in CLI, Web UI, REST, and an agent tool
- Shows each alpha’s gap to the leader
- Supports a subset filter so only selected alphas are evaluated
4. Opt-in local data cache
Backtests can now avoid repeated downloads through the VIBE_TRADING_DATA_CACHE switch. The cache stores settled historical bars under ~/.vibe-trading/cache for seven sources: tushare, okx, ccxt, akshare, mootdx, yfinance, and futu.
That helps when you rerun long-horizon or cross-market tests and want to skip rate limits or slow fetches. The cache is off by default, and the project avoids caching ranges that end today so incomplete bars do not get stored as final data.
- Byte-identical round trips between cached and freshly fetched frames
- Batch and connection loaders can skip network work on a full cache hit
- Cache lives in the user home directory, not the repo
5. Connector-first broker access
The trading layer now starts from selectable connector profiles instead of a hard split between broker and live paths. Vibe-Trading supports direct SDK and MCP routes across brokers, with read-only account, positions, orders, quote, and history access plus paper-account order placement where the broker allows it.
Some connectors also support bounded live placement behind a mandate, filesystem kill switch, pre-trade gate, and audit ledger. That makes the execution model much clearer: paper, read-only, and live are defined by the connector itself, not by a generic mode flag.
- Includes IBKR local read-only TWS / IB Gateway support
- Includes Robinhood Agentic Trading via OAuth and mandate controls
- Recent additions include Tiger, Longbridge, Alpaca, OKX, Binance, Futu, Dhan, and Shoonya
How to decide
Pick the CLI refresh if you want a better daily interface for agent runs. Pick live swarm status if you care about multi-agent coordination. Pick alpha compare if your focus is signal research, and pick the data cache if backtest speed and provider limits are your bottlenecks.
If your goal is actual execution, the connector-first broker layer is the main reason to pay attention. It is the broadest change in the repo, and it defines how Vibe-Trading moves from research output to broker-aware action.
// Related Articles
- [IND]
OpenAI式天价IPO泡沫大于现实,美股牛市还没到终点
- [IND]
Anthropic is right: advanced AI needs a real pause mechanism
- [IND]
Cloudflare’s AI-agent bet is changing the stock case
- [IND]
D.C. splits AI and crypto oversight paths
- [IND]
Microsoft and OpenAI’s split is already visible
- [IND]
Oracle OKE’s Kubernetes support schedule, explained