节省每一个 token。
赋能你的团队。

通过本地优化削减 40-70% 的 AI token 开销——再用 Terse Cloud 让全团队掌握完整用量可视化。按开发者、项目和工具分析 token 开销。

Try Live Demo macOS Windows VS Code Chrome Terse Cloud Terse Docs NEW Join our Slack

✓ 30-day free trial · $0 due today · no card for the demo · cancel anytime

首次启动 — macOS 安全步骤（仅需一次）

macOS 默认阻止未签名的应用程序。请选择适合您的方法：

A 系统设置 — 最简单，无需终端
1. 打开 Terse — 在安全警告上点击好
2. 打开系统设置 → 隐私与安全性
3. 滚动到安全性部分 — 点击 Terse 旁边的“仍要打开”
4. 在弹窗中确认 — 完成。此后 Terse 可正常启动。
B 右键点击打开 — 适用于 macOS Ventura 及更早版本
1. 在 Finder 中，右键点击（或 Control+点击）Terse.app
2. 从菜单中选择打开
3. 在弹出的对话框中点击打开
注意：macOS Sequoia（15+）已移除此选项 — 请使用方法 A。
C 终端命令 — 一条命令，适用所有版本
将 Terse 拖到/Applications，然后在终端中粘贴以下命令：
xattr -cr /Applications/Terse.app && /Applications/Terse.app/Contents/MacOS/terse 2>/dev/null &
OpenClaw — ChatAgent
claude-sonnet-4-6
已连接
0 轮 0 输入 0 输出 0 缓存 $0.000 0 已节省
Google Chrome — ChatGPTBrowser
就绪
VS Code — Cursor ChatEditor
main.js
optimizer.js
capture.js
preload.js
1 const optimizer = new PromptOptimizer();
2 const result = optimizer.optimize(text);
3 console.log(result.stats);
Cursor Chat · Aggr Mode
就绪
Terminal — Claude CodeAgent
检测到 Claude Code 连接
0轮
0输入
0输出
$0成本
Terse — Session ManagerMulti-App
Google Chrome
ChatGPT — claude.ai
ACTIVE
VS Code
terse-project — main.js
Cursor
my-project — index.ts
+ 添加会话
4.8 · loved by developers
40–70%
average token savings
100%
on-device · zero cloud
Works with
Claude Code · Cursor · ChatGPT · Copilot · Aider

Terse is a macOS & iOS token optimization tool that reduces AI prompt costs by 40-70% through 35+ intelligent compression techniques — including git diff compression, prompt caching awareness, conversation history summarization, model-specific optimization, duplicate tool call detection, redundant file read flagging, and real-time per-turn cost tracking across Claude Code, Cursor, ChatGPT, OpenClaw, and Aider. Built on research from LLMLingua (EMNLP 2023), Norvig spelling correction, and selective context pruning.

Updated March 2026 · v1.2.0

C
ChatGPT
CC
Claude Code
OC
OpenClaw
Cu
Cursor
Ai
Aider
VS
VS Code
Sa
Safari
Ch
Chrome
Cl
Claude.ai
Gm
Gemini
Wi
Windsurf
Cp
Copilot
C
ChatGPT
CC
Claude Code
OC
OpenClaw
Cu
Cursor
Ai
Aider
VS
VS Code
Sa
Safari
Ch
Chrome
Cl
Claude.ai
Gm
Gemini
Wi
Windsurf
Cp
Copilot

优化每一条消息

在提示和 Agent 命令到达 API 之前拦截——去除填充词、修正拼写、缩短短语、压缩冗长文本。减少 40-80% 的输入 token。

监控 Agent 会话

自动检测 Claude Code、OpenClaw、Cursor 和 Aider。跟踪输入/输出 token、缓存效率、工具调用开销和每轮成本。

消除浪费的 token

检测重复工具调用，标记多次读取的文件，估算可压缩的工具结果，捕获导致高成本重试循环的拼写错误。

Git diff compression

Detects git diff output and compresses to only changed lines + 1 line of context. Typical 70% reduction — critical for Claude Code and Cursor workflows where diffs dominate tool output.

Prompt caching awareness

Detects stable prefixes and repeated content blocks that benefit from prompt caching — the #1 cost reduction technique in 2026. Saves 50-90% on repeated context across agent turns.

Conversation history summarization

Auto-summarizes old turns in long agent sessions: "User: A B C, Assistant: X Y Z" → "Earlier: Discussed A, concluded X." Massive savings in multi-turn workflows — up to 80% on history context.

Model-specific optimization

Claude handles implicit reasoning — Terse compresses more aggressively. GPT-4 needs explicit instructions — Terse preserves structure. Automatic model detection adapts compression strategy per provider.

兼容
Agent Token 优化

每一轮自动优化。

单个 Agent 任务消耗的 token 可达聊天消息的 50 倍。Terse 在每一轮运行 5 种并行优化策略。

  • 在用户消息到达 API 之前进行压缩
  • 检测重复工具调用——相同工具 + 输入 = 浪费的 token
  • 标记多次读取的文件（每次重读浪费约 800 token）
  • 估算可压缩的工具结果（Read: 60%，Grep: 40%）
  • 跟踪未使用的工具——每次调用中每个未使用工具约 300 token 开销
  • 根据会话模式生成 CLAUDE.md 规则以实现未来节省
Claude CodeOpenClawAiderCursor Agent
Agent Session — Live Optimization
1 You type in agent session
 
 
2 Terse optimizes — compress + dedup + fix
1Typo Correction
2Context Dedup
3Filler Removal
4Prompt Compression
5History Trimming
6Imperative Rewrite
7Final Cleanup
3 Optimized prompt auto-sent to agent
 
 
4 Agent runs — Terse tracks everything
⚙ 
⚙ 
✎ 
▶ 
✓ 
5 Cumulative savings — per-turn cost breakdown
—Turn
—Input tok
—Output tok
—Cache
—Tools
—Tok saved
—Typos
—Cost
 
秒级上手

下载，点击连接。
一切开箱即用。

无需配置文件，无需粘贴 API 密钥，无需终端命令。打开 Terse，看到正在运行的 Agent，点击连接——优化立即开始。

  • 通过进程扫描自动检测 Claude Code、Cursor Agent、Aider、OpenClaw
  • 一键连接——下一条消息即开始优化
  • 实时弹出栏显示每轮节省情况
  • 兼容现有会话——无需重启
Claude CodeCursor AgentAiderOpenClaw
Terse — Connect Agent
检测到 Claude Code — 监控此会话？ 连接
0Turns
0Tok saved
$0Cost
0%Cache hit
为什么选择 Terse

Terse 对比 RTK 及其他
Token 工具库

RTK（Reduce Token Kit）等类似工具库仅提供基本的提示词精简功能。Terse 远不止于此——支持智能体感知优化、实时监控和输出压缩，这些都是 RTK 不具备的能力。

功能
Terse
RTK / Others
优化技术
35+
5-10
工具输出压缩
✓ RTK-style + more
Basic
Git diff 压缩
✓ 70% reduction
✗
提示词缓存感知
✓ 50-90% on repeats
✗
智能体会话监控
✓ Real-time
✗
重复工具调用检测
✓
✗
历史记录摘要
✓ Auto-compress
✗
模型专属优化
✓ Claude / GPT modes
✗
多语言支持
✓ 11 languages
English only
拼写检查管道
✓ Dict + Norvig + macOS
Basic / None
代码保护
✓ Backtick-aware
Partial
Token 交易市场
✓ Buy/sell tokens
✗
本地处理 / 隐私保护
✓ 100% local
Varies

RTK = Reduce Token Kit。对比基于截至 2026 年 3 月的公开功能文档。

随处可用

在任何应用上自动运行。

将 Terse 连接到 Chrome、Cursor、VS Code 或任何终端——自动检测文本框和 Agent 会话，无需安装插件。

  • 自动检测 Agent 进程——Claude Code、OpenClaw、Cursor、Aider
  • 7 阶段流水线在每个提示和 Agent 命令上运行
  • 代码块、URL、行内代码全部受保护 · 本地运行 · 零延迟
Pipeline — Live
1拼写校正
2空白处理
3模式优化
4冗余消除
5NLP 分析
6激进压缩
7最终清理
实时 Token 优化

每条消息实时优化。

在你输入时，Terse 实时重写你的提示——修正拼写、去除填充词、压缩冗长表达。

  • 在发送前重写提示和 Agent 命令
  • 上下文感知的拼写校正
  • 优化 Agent 工具调用、文件读取和上下文传递
  • 安全：跳过全大写、首字母大写和代码 token
Token Optimization — Live
TYPOS Dict
Norvig
Context
macOS Spellcheck
20+ 种技术

找到每一个浪费的 token。

填充词去除、疑问句转祈使句、去重、电报式压缩——每种技术针对不同的浪费来源。

  • 130+ 条短语缩减规则，适用于提示和 Agent 消息
  • 语义去重——捕获 Agent 轮次间重复的上下文
  • 工具结果压缩——标记大型 Read/Grep 结果以供裁剪
Techniques — Live

冗余读取检测

疑问句 → 祈使句

重复工具调用

语义去重

填充词去除

未使用工具开销

三种模式

你控制节省多少。

不同场景需要不同级别。Soft 适合谨慎提示，Normal 适合日常聊天，Aggressive 适合 Agent 会话。

  • Soft：仅拼写修正 + 空白处理。适合关键提示。
  • Normal：去除填充词、犹豫和元语言。最适合聊天。
  • Aggressive：最大压缩 + 电报风格。专为 Agent 会话设计。
Mode Comparison
轻柔
标准
激进
"I was just wondering if you could perhaps help me understand how to implement a binary search tree in Python please?"
22 tok
0%
Agent 监控

查看 Agent 的一切操作。

Terse 通过进程扫描自动检测 Claude Code、OpenClaw、Aider 和 Cursor Agent。实时追踪会话日志，内置模型感知定价。

  • 实时 token 跟踪：输入、输出、缓存读取、上下文填充百分比
  • 模型感知成本估算——内置 Opus/Sonnet/Haiku 定价
  • 重复工具调用检测 + 冗余文件读取告警
  • 未使用工具开销跟踪（每次调用中每个未使用工具约 300 token）
  • 上下文填充度指示器——60% 时警告，85% 时告警
  • 根据会话模式自动生成 CLAUDE.md 优化规则
Claude CodeOpenClawAiderCursor Agent
Agent Monitor — Live
检测到 Claude Code——监控会话？连接
0Turns
0Input
0Output
$0Cost
0Cache
0Tools
0Typos
0sDuration
Prompt savings—
Model:claude-opus-4-6Streaming
自动模型路由

Opus 计费。
Sonnet 响应。

Terse 在 7860 端口运行本地代理，拦截每个 API 请求。简单任务——短提示词、查询、编辑——自动从 Opus（$15/MTok）路由到 Sonnet（$3/MTok）。复杂任务保留 Opus。无需修改任何代码，费用减少 80%。

  • 复杂度评分——短提示词、查询、编辑 → Sonnet
  • 架构设计、安全审查、深度重构 → 保留 Opus
  • 透明无感——Claude Code、Cursor、Codex 无需任何改动
Claude CodeCursorCodexAny OpenAI-compat client
Terse Proxy — Live Routingport 7860
0Requests
0Routed
$0Saved
0%Route rate
性能测试

基于真实会话测试。

在真实的 ChatGPT 提示、Claude Code Agent 会话和多轮 Agent 工作流上测试。简洁的技术提示原样通过，冗长的提示和 Agent 消息可减少 40-70%。

  • 覆盖手动提示、Agent 对话轮次及工具调用的基准测试
  • 简洁提示正确返回 0%——不会产生误改
  • 节省可累积：5 轮 Agent 会话可节省 200-400+ token
Benchmarks — Aggressive Mode
Agent: mixed typos + filler
-64%
Agent: verbose debug prompt
-60%
Claude Code: typo-heavy
-51%
OpenClaw: chatty request
-46%
Tool overhead (unused)
-35%
Repeated context (dedup)
-28%
Clean technical
0%
看看效果

真实输出，真实节省。

在实际提示和 Agent 命令上的逐行对比。

find . -name "*.rs"
cargo test
git diff
git log
verbose prompt
agent debug
find . -name "*.rs" ~276 tokens 
./target/debug/build/serde_core-.../out/private.rs
./target/debug/build/libsqlite3-sys-.../out/bindgen.rs
./target/debug/build/serde-.../out/private.rs
./src/ls.rs
./src/local_llm.rs
./src/learn/detector.rs
./src/learn/report.rs
./src/learn/mod.rs
./src/discover/registry.rs
./src/discover/provider.rs
./src/discover/report.rs
./src/discover/mod.rs
./src/wget_cmd.rs
./src/npm_cmd.rs
./src/cargo_cmd.rs
./src/ccusage.rs
./src/config.rs
./src/lint_cmd.rs
./src/curl_cmd.rs
./src/prisma_cmd.rs
./src/cc_economics.rs
./src/find_cmd.rs
./src/gain.rs
./src/git.rs
... 49 files total
Terse optimized ~149 tokens -46% 
49F 4D:

src/ cargo_cmd.rs cc_economics.rs
  ccusage.rs config.rs container.rs
  curl_cmd.rs deps.rs diff_cmd.rs
  display_helpers.rs env_cmd.rs
  filter.rs find_cmd.rs gain.rs
  gh_cmd.rs git.rs grep_cmd.rs
  init.rs json_cmd.rs lint_cmd.rs
  local_llm.rs log_cmd.rs ls.rs
  main.rs ...
src/discover/ mod.rs provider.rs
  registry.rs report.rs
src/learn/ detector.rs mod.rs report.rs
src/parser/ error.rs formatter.rs
  mod.rs types.rs
cargo test ~4,823 tokens 
   Compiling serde v1.0.210
   Compiling serde_json v1.0.128
   Compiling tokio v1.40.0
   Compiling reqwest v0.12.7
   Compiling my-project v0.1.0 (/Users/dev/project)
    Finished `test` profile [unoptimized + debuginfo] target(s) in 45.23s
     Running unittests src/main.rs (target/debug/deps/my_project-a1b2c3d4)

running 262 tests
test config::tests::test_default_config ... ok
test config::tests::test_parse_env ... ok
test config::tests::test_merge_configs ... ok
test git::tests::test_parse_diff ... ok
test git::tests::test_status_parse ... ok
... 257 more tests
test result: ok. 262 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 3.41s
Terse optimized ~38 tokens -99% 
cargo test: 262 passed, 0 failed
Built 5 crates in 45.23s
All tests green.
git diff ~1,240 tokens 
diff --git a/src/optimizer.js b/src/optimizer.js
index 3a4b5c6..7d8e9f0 100644
--- a/src/optimizer.js
+++ b/src/optimizer.js
@@ -142,8 +142,12 @@ class PromptOptimizer {
     // Remove filler words from text
-    const fillers = ['just', 'really', 'very', 'actually'];
+    const fillers = ['just', 'really', 'very', 'actually',
+      'basically', 'literally', 'honestly', 'perhaps',
+      'maybe', 'probably', 'simply'];
     for (const f of fillers) {
-      text = text.replace(new RegExp(`\\b${f}\\b`, 'gi'), '');
+      text = text.replace(
+        new RegExp(`\\b${f}\\b\\s*`, 'gi'), ''
+      );
     }
@@ -165,3 +169,8 @@ class PromptOptimizer {
+  telegraphCompress(text) {
+    return text.replace(/\b(the|a|an)\b/gi, '')
+               .replace(/\s{2,}/g, ' ')
+               .trim();
+  }
Terse optimized ~310 tokens -75% 
src/optimizer.js: 2 hunks, +13 -3

L142: expanded fillers list
  (7 new: basically, literally, honestly,
   perhaps, maybe, probably, simply)
  + regex now strips trailing whitespace

L169: added telegraphCompress()
  strips articles (the/a/an), collapses spaces
git log --oneline -20 ~520 tokens 
a1b2c3d Fix authentication module token refresh handling
e4f5g6h Update README with new API documentation
i7j8k9l Refactor optimizer pipeline for better performance
m0n1o2p Add spellcheck integration with macOS NSSpellChecker
q3r4s5t Fix duplicate tool call detection edge case
u6v7w8x Update dependencies: electron 40, node 22
y9z0a1b Add aggressive mode telegraph compression
c2d3e4f Fix session manager reconnection bug
g5h6i7j Implement CLAUDE.md rule generation from patterns
k8l9m0n Add context fill meter warning at 60% and 85%
o1p2q3r Fix AX read fallback for Electron editors
s4t5u6v Add Jaccard similarity dedup for agent turns
w7x8y9z Refactor capture module for multi-window support
a0b1c2d Fix popup window focus stealing on macOS
e3f4g5h Add model-aware pricing for Opus/Sonnet/Haiku
i6j7k8l Update landing page hero animation
m9n0o1p Fix shell hook installation path resolution
q2r3s4t Add RTK-style output compression techniques
u5v6w7x Fix cache efficiency tracking in agent monitor
y8z9a0b Initial commit
Terse optimized ~185 tokens -64% 
20 commits, 3 authors

Recent: auth token fix, README update,
optimizer refactor, spellcheck integration,
dedup edge case fix

Themes: optimizer (4), agent monitor (3),
bug fixes (6), features (5), docs (2)
User prompt ~52 tokens 
I was just wondering if you could perhaps maybe help me understand how to implement a binary search tree in Python? I'm not really sure about the best approach to take here and I would really appreciate any guidance you could provide please.
Terse optimized ~11 tokens -79% 
Implement binary search tree in Python. Show best approach.
Agent debug prompt ~68 tokens 
I don't know if this makes sense but the authetication module is broken again. Could you maybe look into it and try to figure out why the tokne refresh isn't working properly? Like I mentioned earlier, the refresh endpoint keeps returning 401 errors and I really need this fixed as soon as possible please.
Terse optimized ~18 tokens -74% 
Fix authentication module: token refresh returns 401. Debug refresh endpoint. Priority: high.

没有任何 AI 工具提供无限使用。

即使每月 200 美元的方案也有上限。Terse 压缩提示让额度用得更久——Terse Cloud 让你的团队清晰掌握每个 token 的去向。

一次典型的 2 小时 AI Agent 编程会话：
0
条 CLI 命令执行
0
个 token 的提示 + CLI 噪声
0
使用 Terse 后（减少 89%）

没有 Terse，CLI 输出和冗长提示就能撑爆 200K 上下文窗口。

每款工具都有限制

Terse 让每个方案都更耐用。

无论团队使用哪款 AI 工具，token 成本都会快速累积。Terse 压缩输入内容——Terse Cloud 提供团队级分析，按开发者、项目和工具追踪开销。

Claude Code Terminal
价格$20 — $200/mo
限制~45 msgs/5h (Pro), 5-20x on Max
上下文200K tokens
会话时长约 3 倍
Even Max $200/mo (20x Pro) has weekly caps (240-480h). Quota resets every 5h. Terse compresses prompts and CLI outputs by avg 89%, so each message carries less noise and your quota stretches ~3x.
Cursor IDE
价格$20 — $200/mo
限制$20 credits/mo (Pro), ~225 Claude reqs
上下文Up to 200K (Max mode)
额度使用效率约 2 倍
Even Ultra $200/mo (20x credits) is capped. Each request consumes credits based on model — Claude burns 2.4x faster than Gemini. Terse compresses prompts and CLI outputs so each request starts cleaner.
OpenAI Codex Agent
价格$20/mo (Plus) — $200/mo (Pro)
限制30-1,500 msgs/5h by plan
上下文192K tokens
更多迭代次数
Included with ChatGPT plans. Pro $200/mo caps at 1,500 msgs/5h. The agent runs commands autonomously — each output eats your cap. Terse compresses them for more iterations per window.
Windsurf IDE
价格$15 — $60/mo
限制500 credits/mo (Pro)
上下文200K tokens
额度持续约 2 倍
Enterprise $60/user gets 1,000 credits/mo — still capped. Cascade consumes credits per prompt. Terse compresses prompts and CLI outputs so each interaction uses fewer tokens, stretching your credits.
Gemini CLI Terminal
价格Free — pay-per-token
限制1,000 req/day, 60 req/min (free)
上下文1M tokens
token 账单减少约 70%
Free tier is generous (1,000 req/day) but still rate-limited. Beyond that, you pay per token. Terse compresses prompts and CLI outputs by avg 89%, cutting your bill or freeing rate limit headroom.
Aider Terminal
价格Free + API costs ($5-300+/mo)
限制Per API provider
上下文Per model (up to 200K)
API 成本减少约 70%
BYO API key — you pay per token to OpenAI, Anthropic, etc. Terse compresses every prompt and command output before it reaches the model, directly cutting your API bill by ~70% on verbose workflows.
GitHub Copilot IDE
价格Free — $39/mo (Pro+)
限制50-1,500 premium req/mo
上下文Per model (up to 200K)
更好的上下文质量
Enterprise $39/user: 1,000 premium req/mo. Base completions are unlimited, but Chat and the coding agent have caps. Terse keeps terminal output lean so premium requests carry more useful context.
Cline / Roo VS Code
价格Free + API costs ($0-500+/mo)
限制Per API provider
上下文Per model (up to 200K)
API 成本减少约 70%
No tool-side limit, but your API provider caps apply. Heavy users report $200-500+/mo. Terse compresses every output by avg 89%, directly cutting your bill and reducing context overflow.

价格信息已于 2026 年 2 月核实。限制因使用量和方案而异。

Terse Pals

Your companion through every session

Pick a pal that celebrates every token you save — it reacts to each tool call, eats your savings, and keeps you company through long coding sessions.

20 pals available  ·  Click to poke  ·  Unlock new ones as you save tokens

……深受开发者喜爱

工程师和 AI 重度用户正在降低成本，同时全面掌控 token 使用情况。

M
Marcus Chen
@marcuschen_dev
用 @Terse_App 配合 Claude Code 一周了。Agent 会话的 token 用量下降了约 40%。仅拼写检查就帮我避免了高成本的纠错循环。
Claude Code
S
Sarah Kim
@sarahk_ai
我习惯性地输入冗长的 ChatGPT 提示。Terse 实时捕获我所有的填充词和犹豫表达。平均减少 60% 的 token。
ChatGPT
J
Jake Ortiz
@jakeortiz
仅 Agent 监控就物超所值。我可以看到每轮的输入/输出/缓存、工具调用成本，以及哪些会话消耗最多 token。
Cursor Agent
A
Amara Patel
@amara_codes
100% 本地运行。无 API 调用，无云端。作为处理敏感代码库的人，这是我唯一信任的 token 优化器。
Privacy
R
Ravi Nguyen
@ravi_ng
配置了 OpenClaw + Terse，API 账单立刻降低。自动模式在发送前重写提示，监控器跟踪每轮成本。
OpenClaw
E
Elena Vasquez
@elena_v
三种模式非常完美。Soft 用于重要提示，Aggressive 用于快速随手问题。Terse 适应我的工作方式。
3 Modes
D
Daniel Park
@dpark_dev
我每月有 $40 的 Claude API 额度闲置。在 Token 交易所上架了我的密钥——现在每月从其他开发者那里赚约 $25。
Token Exchange
L
Lisa Chen
@lisachen_ai
作为学生，全价购买 Claude API 很心痛。Token 交易所让我以 6 折获得 Sonnet。
Token Exchange

基于学术研究。

基于 LLMLingua、Norvig 拼写校正、选择性上下文剪枝以及真实 Agent 会话分析。

0
种优化策略
0
项 token 压缩技术
0
家 API 服务商支持
0
% 交易所最高折扣
Developer API

Terse API

Integrate token optimization directly into your vibe coding project. One API call — 30–60% fewer tokens, same meaning, same results.

Read the docs GitHub examples
🔑

1. Get your API key

Sign in and generate a tsk_... developer key from your dashboard. Free tier included.

⚡

2. Optimize before sending

POST your prompt to /api/v1/optimize — get back a trimmed version with token count saved.

📡

3. Scan your codebase

Use /api/v1/scan to find every LLM call site in your project and see where tokens are wasted.

// Install: (no package needed — just HTTP)
const response = await fetch('https://www.terseai.org/api/v1/optimize', {
  method: 'POST',
  headers: {
    'Authorization': `Bearer ${process.env.TERSE_API_KEY}`,
    'Content-Type': 'application/json',
  },
  body: JSON.stringify({
    text: userPrompt,   // your prompt string
    mode: 'normal',     // 'soft' | 'normal' | 'aggressive'
  }),
});
const { optimized, tokens_saved, reduction_pct } = await response.json();
// → use `optimized` as the prompt for your LLM call

API Pricing

Free tier included with every account. No credit card for free. Upgrade for higher limits and batch access.

Free
$0/mo

No credit card needed.

  • 500K tokens / month
  • 60 requests / minute
  • All 3 compression modes
  • Community support
Most Popular
Pro
$29/mo

Cancel anytime. No trial.

  • 50M tokens / month
  • 600 requests / minute
  • Batch compression
  • Usage analytics dashboard
  • Email support
Enterprise
Custom

Teams & high-volume use.

  • Unlimited tokens
  • Custom rate limits
  • Dedicated endpoints
  • SLA + priority support
  • Team management + audit logs
Contact us →
Modes
soft Typo fix + whitespace + safe phrase shortening. Meaning 100% preserved.
normal + filler removal, hedging, politeness, question→imperative conversion.
aggressive + abbreviations, markdown strip, article removal, telegraph compression.
Endpoints
POST /api/v1/optimize — optimize a prompt
POST /api/v1/scan — scan code for LLM calls
POST /api/v1/keys — create API key
GET  /api/v1/keys — list your keys
POST /api/v1/projects — publish your project
GET  /api/v1/projects — browse showcase
Platform

Vibe Coding Projects

Built with Terse API? Publish your project to get traffic, users, and visibility in the vibe coding developer community.

Browse all projects →
🤖
ClaudeFlow
Multi-agent workflow builder
247

Orchestrates Claude agents for complex tasks. Terse API cuts context window usage by 41% per turn, enabling longer sessions without hitting limits.

Node.js Agents
~38K tokens saved/mo
💬
PromptKit
Open-source prompt library
189

A library of 200+ production-ready prompts. Uses Terse API to auto-compress each prompt at runtime, reducing developer API bills by $80/mo on average.

TypeScript Prompts
~21K tokens saved/mo
GitHub ↗
🐍
PyAgentKit
Python agent scaffolding
134

Batteries-included Python scaffold for building LLM agents. Terse integration is built-in — every outbound message is auto-optimized before the API call.

Python Scaffold
~29K tokens saved/mo
GitHub ↗

Have a project using Terse API?

Teams

Terse Cloud

Visibility and control over your team's AI coding costs. See who's using what, how much is saved, and where tokens are wasted — across every tool.

📊

Token analytics

Dashboard by developer, project, and tool — Mac, Windows, Chrome, VS Code, iOS.

💰

Team savings report

"Your team saved $4,200 this month" — with per-developer and per-project breakdowns.

🔔

Rate-limit alerts

Smart notifications when a developer is approaching heavy usage thresholds.

🔑

Team token control

Distribute API tokens, add or remove developers, and audit all activity in one place.

Open team dashboard →

Free for open source. Teams from $15 / developer / month.

价格

简单透明的方案

按月方案含 30 天免费试用 — 今天 $0。也可选择按周或按季付费。

Pro
$4.99/mo
或 $1.99/周 · $12/季（约 $4/月 — 省 20%）

面向每日使用 Agent 的开发者。无限提示优化，多会话监控。

30 天免费试用——随时取消
  • 无限次优化
  • 3 个连接会话
  • 2 台设备
  • 全部 3 种优化模式
  • Agent 监控 + 重复检测
  • 自动替换和发送模式
  • CLAUDE.md 规则生成
Premium
$99/mo

面向团队和重度用户。全部功能无限制，优先支持。

30 天免费试用——随时取消
  • 无限次优化
  • 无限连接会话
  • 无限设备
  • 全部 3 种优化模式
  • 完整 Agent 分析 + 规则生成
  • 自动替换和发送模式
  • 优先支持
常见问题

常见问题解答

关于 token 优化以及 Terse 如何帮你省钱，你需要了解的一切。

什么是 token 优化？
Token 优化在不丢失语义的前提下减少 AI 提示中的 token 数量。Terse 使用 20 多种技术将用量降低 40-70%，直接降低 AI API 成本。
Terse 能节省多少？
冗长提示节省 40-70%，CLI 输出最高节省 89%。典型的 2 小时会话从约 21 万 token 降至约 2.3 万。
哪些 AI 工具可以配合 Terse 使用？
Claude Code、Cursor、OpenClaw、Aider 及所有终端 AI Agent。也支持通过 macOS 辅助功能 API 使用的浏览器工具。
提示压缩是怎么实现的？
7 阶段流水线：拼写校正、空白规范化、模式优化、冗余消除、NLP 分析、电报式压缩和清理。
会降低输出质量吗？
不会。它只去除噪声——填充词、犹豫表达、拼写错误——不改变意图。研究表明压缩后的提示能保持甚至提升质量。
和提示工程有什么区别？
提示工程是编写更好的指令，token 优化是通过去除冗余来降低成本。Terse 自动完成优化工作。
Terse 免费吗？
两种方案均包含 30 天免费试用。Pro 版 $4.99/月，Premium 版 $99/月，随时可取消。
Token 如何影响成本？
AI 模型按 token 计费（约 4 个字符为 1 个 token）。单次 Agent 会话可消耗 20 万以上 token，费用 $3-$15。
什么是 Token 交易所？
一个交易闲置 AI API token 的市场。卖家折价出售密钥，买家获得更便宜的访问。
如何买卖 token？
在 terseai.org/marketplace 登录。卖出：粘贴密钥，设定折扣。买入：充值，生成 API 密钥。
News & Insights

Stay ahead on AI cost optimization.

Research, guides, and analysis on reducing LLM token costs — from prompt caching to model routing to agent session efficiency.

Cost Optimization

Prompt Caching in 2026: How to Cut Claude Code Costs by 73%

Prompt caching is now the single highest-ROI technique for reducing AI agent costs. Cached tokens cost 10× less than standard input tokens — here's how to structure every session for maximum cache hit rate.

8 min read
Agent Optimization

Git Diff Compression: Reclaim 70% of Your Agent Context Window

When Claude Code runs git diff, the output can consume 30–60K tokens per turn. Compressing to changed lines + 1 line of context is the fastest context win in any agent workflow.

6 min read
Research

The Real Cost of Duplicate Tool Calls in AI Agent Sessions

Analysis of 500 real Claude Code sessions found 34% of Read tool calls were redundant. At $15/MTok for Opus, each repeated file read wastes ~$0.006 — small per turn, painful at scale.

5 min read
Model Selection

Model Routing in 2026: When Sonnet Outperforms Opus at 1/5th the Cost

Not every agent task needs Claude Opus. Single-file edits, test generation, and short lookups perform identically on Sonnet — saving teams $200+/month with zero quality loss on routine tasks.

7 min read
Technique Comparison

LLMLingua vs. Rule-Based Compression: A Developer's Benchmark

LLMLingua uses perplexity scoring to drop tokens at inference time. Deterministic rule pipelines work at the prompt layer before the API call. We tested both on 1,000 real developer prompts.

9 min read
Research

Why Typos in Agent Prompts Cost More Than You Think

A single typo in an agent prompt can trigger a clarification turn, adding 200–500 tokens of overhead. Typo-caused retries account for 8–15% of total token spend across a typical week of Claude Code usage.

4 min read
Weekly Newsletter

Get the token savings playbook
in your inbox.

Practical techniques — prompt caching, model routing, context compression — delivered weekly. No fluff.

Join developers cutting AI costs every week. Unsubscribe anytime.

别再浪费 token 和金钱。

优化每一条提示，监控每一个会话。用 Terse Cloud 为全团队提供可视化——按开发者、项目和工具分析。

macOS Windows VS Code Chrome Terse Cloud Join our Slack

首次启动 — macOS 安全步骤（仅需一次）

macOS 默认阻止未签名的应用程序。请选择适合您的方法：

A 系统设置 — 最简单，无需终端
1. 打开 Terse — 在安全警告上点击好
2. 打开系统设置 → 隐私与安全性
3. 滚动到安全性部分 — 点击 Terse 旁边的“仍要打开”
4. 在弹窗中确认 — 完成。此后 Terse 可正常启动。
B 右键点击打开 — 适用于 macOS Ventura 及更早版本
1. 在 Finder 中，右键点击（或 Control+点击）Terse.app
2. 从菜单中选择打开
3. 在弹出的对话框中点击打开
注意：macOS Sequoia（15+）已移除此选项 — 请使用方法 A。
C 终端命令 — 一条命令，适用所有版本
将 Terse 拖到/Applications，然后在终端中粘贴以下命令：
xattr -cr /Applications/Terse.app && /Applications/Terse.app/Contents/MacOS/terse 2>/dev/null &
100% 本地运行 零延迟

Terse 关键事实

Terse v1.3.3 · 数据核对于 2026-08-28

Token 削减
冗长提示词降低 40-70%,CLI 输出最高 89%
原理
7 段流水线、35+ 项技术,全部本地运行,耗时低于 5 毫秒
监控的智能体
8 个 —— Claude Code、Cursor、OpenAI Codex、GitHub Copilot CLI、Cline、Windsurf、OpenClaw、Aider
支持平台
macOS 与 Windows,另有 Chrome 与 VS Code 扩展
隐私
100% 本地处理。提示词、源码与会话数据不离开本机。
价格
免费版;Pro 每月 $4.99(30 天免费试用);Premium 每月 $99