|
| 1 | +--- |
| 2 | +# / Context: https://ctx.ist |
| 3 | +# ,'`./ do you remember? |
| 4 | +# `.,'\ |
| 5 | +# \ Copyright 2026-present Context contributors. |
| 6 | +# SPDX-License-Identifier: Apache-2.0 |
| 7 | + |
| 8 | +icon: lucide/git-compare |
| 9 | +--- |
| 10 | + |
| 11 | + |
| 12 | + |
| 13 | +## Similar Tools and How `ctx` Differs |
| 14 | + |
| 15 | +There are many tools in the AI ecosystem that touch *parts* of the context |
| 16 | +problem: |
| 17 | + |
| 18 | +* Some manage prompts. |
| 19 | +* Some retrieve data. |
| 20 | +* Some provide runtime context objects. |
| 21 | +* Some offer enterprise platforms. |
| 22 | + |
| 23 | +`ctx` focuses on a different layer entirely. |
| 24 | + |
| 25 | +This page explains where `ctx` fits, and where it **intentionally** does not. |
| 26 | + |
| 27 | +--- |
| 28 | + |
| 29 | +## The Core Distinction |
| 30 | + |
| 31 | +Most tools treat context as **input**. |
| 32 | + |
| 33 | +`ctx` treats context as **infrastructure**. |
| 34 | + |
| 35 | +That single difference explains nearly all of `ctx`'s design choices. |
| 36 | + |
| 37 | +| Question | Most tools | ctx | |
| 38 | +|--------------------------|---------------------------|------------------| |
| 39 | +| Where does context live? | In prompts or APIs | In files | |
| 40 | +| How long does it last? | One request / one session | Across time | |
| 41 | +| Who can read it? | The model | Humans and tools | |
| 42 | +| How is it updated? | Implicitly | Explicitly | |
| 43 | +| Is it inspectable? | Rarely | Always | |
| 44 | + |
| 45 | +--- |
| 46 | + |
| 47 | +## Prompt Management Tools |
| 48 | + |
| 49 | +Examples include: |
| 50 | + |
| 51 | +* prompt templates |
| 52 | +* reusable system prompts |
| 53 | +* prompt libraries |
| 54 | +* prompt versioning tools |
| 55 | + |
| 56 | +These tools help you *start* a session. |
| 57 | + |
| 58 | +They do not help you *continue* one. |
| 59 | + |
| 60 | +### How they differ from `ctx` |
| 61 | + |
| 62 | +Prompt tools: |
| 63 | +* inject text at session start |
| 64 | +* are ephemeral by design |
| 65 | +* do not evolve with the project |
| 66 | + |
| 67 | +`ctx`: |
| 68 | +* persists knowledge over time |
| 69 | +* accumulates decisions and learnings |
| 70 | +* makes the context part of the repository itself |
| 71 | + |
| 72 | +Prompt tooling and `ctx` are complementary; not competing. |
| 73 | +Yet, they operate at different layers. |
| 74 | + |
| 75 | +--- |
| 76 | + |
| 77 | +## Retrieval-Augmented Generation (RAG) |
| 78 | + |
| 79 | +RAG systems typically: |
| 80 | + |
| 81 | +* index documents |
| 82 | +* embed text |
| 83 | +* retrieve chunks dynamically at runtime |
| 84 | + |
| 85 | +They are excellent for: |
| 86 | +* large knowledge bases |
| 87 | +* static documentation |
| 88 | +* reference material |
| 89 | + |
| 90 | +### How they differ from `ctx` |
| 91 | + |
| 92 | +RAG answers questions like: |
| 93 | + |
| 94 | +> “What information might be relevant right now?” |
| 95 | +
|
| 96 | +`ctx` answers a different question: |
| 97 | + |
| 98 | +> “What have we already decided, learned, or committed to?” |
| 99 | +
|
| 100 | +Here are some key differences: |
| 101 | + |
| 102 | +| RAG | ctx | |
| 103 | +|-----------------------|-----------------------| |
| 104 | +| Statistical relevance | Intentional relevance | |
| 105 | +| Embedding-based | File-based | |
| 106 | +| Opaque retrieval | Explicit structure | |
| 107 | +| Runtime query | Persistent memory | |
| 108 | + |
| 109 | +`ctx` does not replace RAG. |
| 110 | +Instead, it defines a persistent context layer that RAG can optionally augment. |
| 111 | + |
| 112 | +> RAG belongs to the **data plane**; ctx defines the **context control plane**. |
| 113 | +
|
| 114 | +It focuses on **project memory**, not knowledge search. |
| 115 | + |
| 116 | +--- |
| 117 | + |
| 118 | +## Agent Frameworks |
| 119 | + |
| 120 | +Agent frameworks often provide: |
| 121 | + |
| 122 | +* task loops |
| 123 | +* tool orchestration |
| 124 | +* planner/executor patterns |
| 125 | +* autonomous iteration |
| 126 | + |
| 127 | +These systems are powerful, but they typically assume that: |
| 128 | + |
| 129 | +* memory is external |
| 130 | +* context is injected |
| 131 | +* state is transient |
| 132 | + |
| 133 | +### How they differ from `ctx` |
| 134 | + |
| 135 | +Agent frameworks answer: |
| 136 | + |
| 137 | +> “How should the agent act?” |
| 138 | +
|
| 139 | +`ctx` answers: |
| 140 | + |
| 141 | +> “What should the agent remember?” |
| 142 | +
|
| 143 | +Without persistent context, agents tend to: |
| 144 | +* rediscover decisions |
| 145 | +* repeat mistakes |
| 146 | +* lose architectural intent |
| 147 | + |
| 148 | +This is why `ctx` pairs well with loop-based workflows such as the |
| 149 | +[Ralph Wiggum technique](https://ghuntley.com/ralph/): |
| 150 | + |
| 151 | +* **Ralph** provides iteration |
| 152 | +* **ctx** provides continuity |
| 153 | + |
| 154 | +Together, loops become cumulative instead of forgetful. |
| 155 | + |
| 156 | +--- |
| 157 | + |
| 158 | +## SDK-Level Context Objects |
| 159 | + |
| 160 | +Some SDKs expose "context" objects that exist: |
| 161 | + |
| 162 | +* inside a process |
| 163 | +* during a request |
| 164 | +* for the lifetime of a call chain |
| 165 | + |
| 166 | +These are extremely useful and completely different. |
| 167 | + |
| 168 | +SDK context objects: |
| 169 | +* are in-memory |
| 170 | +* disappear when the process ends |
| 171 | +* are not shared across sessions |
| 172 | + |
| 173 | +`ctx`: |
| 174 | +* survives process restarts |
| 175 | +* survives new chats |
| 176 | +* survives new days |
| 177 | + |
| 178 | +They share a name, not a purpose. |
| 179 | + |
| 180 | +--- |
| 181 | + |
| 182 | +## Enterprise Context Platforms |
| 183 | + |
| 184 | +Enterprise platforms often provide: |
| 185 | + |
| 186 | +* centralized context services |
| 187 | +* dashboards |
| 188 | +* access control |
| 189 | +* organizational knowledge layers |
| 190 | + |
| 191 | +These tools are designed for: |
| 192 | +* teams |
| 193 | +* governance |
| 194 | +* compliance |
| 195 | +* managed environments |
| 196 | + |
| 197 | +`ctx` is intentionally: |
| 198 | + |
| 199 | +* local-first |
| 200 | +* file-based |
| 201 | +* dependency-free |
| 202 | +* CLI-driven |
| 203 | +* developer-controlled |
| 204 | + |
| 205 | +It does not require: |
| 206 | +* a server |
| 207 | +* a database |
| 208 | +* an account |
| 209 | +* a SaaS backend |
| 210 | + |
| 211 | +`ctx` optimizes for *individual and small-team workflows* where context should |
| 212 | +live next to code; **not** behind a service boundary. |
| 213 | + |
| 214 | +--- |
| 215 | + |
| 216 | +## When `ctx` Is a Good Fit |
| 217 | + |
| 218 | +`ctx` works best when: |
| 219 | + |
| 220 | +* you want AI work to compound over time |
| 221 | +* architectural decisions matter |
| 222 | +* context must be inspectable |
| 223 | +* humans and AI must share the same source of truth |
| 224 | +* Git history should include *why*, not just *what* |
| 225 | + |
| 226 | +--- |
| 227 | + |
| 228 | +## When `ctx` Is Not the Right Tool |
| 229 | + |
| 230 | +`ctx` is probably not what you want if: |
| 231 | + |
| 232 | +* you only need one-off prompts |
| 233 | +* you rely exclusively on RAG |
| 234 | +* you want autonomous agents without a human-readable state |
| 235 | +* you require centralized enterprise control |
| 236 | +* you want black-box memory systems |
| 237 | + |
| 238 | +These are valid goals; just different ones. |
| 239 | + |
| 240 | +--- |
| 241 | + |
| 242 | +## Summary |
| 243 | + |
| 244 | +Many tools help AI *think*. |
| 245 | + |
| 246 | +`ctx` helps AI *remember*. |
| 247 | + |
| 248 | +**NOT** by storing thoughts, |
| 249 | +**BUT** by preserving intent. |
0 commit comments