# Kestrel vs Bloomberg Terminal: A Terminal for Humans, a Language for Agents (/blog/kestrel-vs-bloomberg-terminal) 2026-07-12 · kestrel.markets ## Answer card Bloomberg Terminal and Kestrel are different categories, so "which is better" is the wrong question. Bloomberg is a vast, integrated cockpit for a human professional: news, chat, fixed income, FX, equities, research. Kestrel is an open-source language plus runtime for agentic trading: it renders markets as text a model can read and fires contingent programs in milliseconds. If a human watches the screen, buy Bloomberg. If an agent does the watching, that is what Kestrel is for. ## Two different readers The distinction that organizes everything below is: **who reads the screen?** Bloomberg's primary reader is a human analyst. Its genius is density and breadth: a single workstation that fuses real-time news, instant messaging with the entire Street, fixed income and rates analytics, FX, commodities, equity research, and issuer filings into one keyboard-driven surface. Decades of coverage depth live behind those function keys. Nothing in this article disputes that Bloomberg is the best tool in the world for its job. Kestrel's primary reader is a model. A language model cannot "watch" a candlestick chart the way a person glances at one; pixels are not a good interface for a token predictor. Kestrel's wager is **the Interface Thesis**: *the two failures of LLM trading are interface, perception and latency, not intelligence.* Models are already smart enough to reason about markets. They have been handed the wrong interface. Kestrel attacks both halves. ### Perception: the chart is in text A Kestrel `VIEW` renders the market as a compact text **Frame**, a chart-in-text sized for a context window. It is a vertical, append-only tape: one row per candle, newest last, quoted as relative moves in basis points versus the prior close, anchored by periodic keyframes so absolute price never drifts. Because the tape only appends, perception cost is **O(new bars), not O(screen)**: KV-cache-friendly, so an agent re-reads only what changed. ``` VIEW tape budget 1200 tokens price(SPX 5m) # relative candles in bps, keyframed velocity(5m) # rate-of-change pane session(HOD LOD VWAP) # anchors ``` *Target grammar; parser parity tracked in the kestrel repo.* *A quote is not a value.* The Frame is engineered so a model can cite what it sees rather than hallucinate a chart it cannot actually look at. ### Latency: the agent is never in the hot path A Kestrel `PLAN` is a standing, bounded-risk contingent program (trigger, actions, bracket, invalidation, TTL). The runtime fires it in milliseconds and wakes the agent in parallel. This is **fire-then-inform**: execution does not wait on a round-trip through an LLM. *The agent is never in the hot path.* ``` IMPORT fade-ladder FROM ./armory/reversion.kestrel USING signal SPX exec SPY 0dte PLAN hod-fade budget 2R ttl +45m regime {chop} WHEN spot crosses above hod AND velocity(5m) >= p95 AND held 120s DO buy 1 ATM P @ lean(bid, fair, 0.5) RELOAD WHEN spot +10bps buy 1 +1 P @ fair-3c TP 2x frac 0.5 @ fair EXIT velocity(5m) <= p50 @ bid INVALIDATE spot > hod +25bps -> halt ``` *Target grammar; parser parity tracked in the kestrel repo.* This is illustrative, not a recommendation: it shows *how an agent could express* a mean-reversion fade, not a strategy anyone should run. You would deploy a template like this into **your** pod and arm it yourself. Two more statement kinds round out the language. `WAKE` is a standing subscription over a trigger algebra: event-driven attention that spends tokens, never risk. `GRADE` is the honest, counterfactual result of a run: contamination-fenced (LLM authors are graded only on date-blinded days after their training cutoff), scored `VS ungated` and `VS null` as counterfactuals in the syntax, because *a backtest is never flattering* and *weights leak what code fences can't stop.* ## The comparison table | Dimension | Bloomberg Terminal | Kestrel | |---|---|---| | Primary reader | Human analyst / trader | A model (agent) | | Perception model | Dense visual GUI, function keys, charts | Text Frame; the chart is in text, O(new bars) | | Agent in the hot path? | N/A; human is the operator | No; fire-then-inform, runtime executes | | Latency floor | Human reaction time to a screen | Runtime fires in milliseconds | | Native interface / MCP | Terminal app + Bloomberg API (B-PIPE, etc.) | HTTP+SSE canonical; TS SDK, CLI, MCP as equal faces | | Agent-native auth (Envelope) | No; human login / enterprise entitlements | Envelope {scope, budget, ceiling, expiry, revocation} | | Machine payment | No; enterprise contract / seat licensing | Agent wallet (Stripe MPP / x402) or human claim-and-fund | | Evaluation honesty | User's own backtesting tools | GRADE: contamination-fenced, VS null / VS ungated | | Provenance | Institutional-grade licensed data | Certified Blotters + Grades, shareable Proof URL | | Live model (custody) | Not a broker for its data seats | BYO broker via OAuth; never custody; certification over custody | | Data licensing | Vast proprietary + licensed feeds | Databento served as derived works + BYO Alpaca | | Activation path | Sales contract, seat provisioning | Proof-before-account: trial capability → sim → 402 offer | | Pricing | Premium per-seat subscription | OSS language; platform is usage/scarcity-priced via 402 Offers | | Best for | Humans covering many asset classes | Agents that perceive and act on markets | Note on the Bloomberg column: where a cell would require internal pricing or product specifics that public materials do not verify, it describes the dimension generically rather than inventing numbers. Bloomberg's exact seat price, API tiers, and entitlement structures are set by Bloomberg and change; treat those cells as directional. ## Where Bloomberg wins Be honest: on breadth, Bloomberg wins decisively, and it is not close. - **Coverage.** News, issuer research, fixed income, rates, FX, commodities, munis, structured products: a single surface spanning asset classes Kestrel does not touch. - **The network.** Bloomberg chat (IB/MSG) is where a large part of institutional flow is negotiated. That is a social and liquidity moat, not a feature you rewrite in a language. - **Human ergonomics.** For an analyst reading filings, building a screen of comparables, or messaging a counterparty, a dense GUI is the *right* interface. Kestrel would be a worse tool for that human. - **Institutional trust.** Decades of licensed data, entitlements, and enterprise integration across a hundred asset classes. Kestrel is deliberately narrow by comparison: an agent's interface to a focused slice, not an analyst's window onto everything. If your job is human research across many markets, Bloomberg is the answer and Kestrel is not a substitute. ## Where Kestrel is NOT the fit Equally honest: Kestrel is narrow on purpose. - **You are the trader, watching a screen.** If a human is making and timing the decisions, Kestrel adds a language between you and the market you do not need. Use a terminal. - **You need broad multi-asset coverage or a dealer network.** Kestrel is focused on agent perception and execution on liquid, generic instruments: not a cross-asset research desk, and not a chat network. - **You want managed money or advice.** Kestrel is impersonal infrastructure: not investment advice, not a recommendation engine. Its platform is BYO-plan and BYO-broker for live, deliberately regulatory-clean: *open judge, sell certification.* - **You want it all today.** The commercial platform (kestrel.markets) is early. Anonymous trial sims, certified Grades, shareable proof URLs, and 402 Offers run today; the free tier needs no signup. ## The division of labor The cleanest way to hold both tools: Bloomberg is the cockpit built for a pilot. Kestrel is a language for when there is no pilot in the seat, when a model is doing the reading and a runtime is doing the pressing. The commercial platform, kestrel.markets, hosts only what stays scarce when intelligence is abundant: capital and trading authority, licensed data, compute and broker access, deterministic execution, and provenance. Brains stay outside (the **external-agent invariant**). Its stance is *host the scarcity, rent the genius*: **sell the airport, not the pilots.** That is a fundamentally different business than selling a seat to a human, which is why "vs" flatters neither product. They serve different readers. ## The citable one-liner Bloomberg built the best terminal for humans; Kestrel is building the interface for the machines that read faster than any human can watch.