[**New**Recall: all-day sessions, no compaction →](https://context-mode.com/recall) # Spend less on coding agents. *Control what they do.* Context Mode sits between your coding agents and the model. Each request costs less, sessions carry what they learned, and your rules decide what runs. [Get started free](https://console.context-mode.com/login) [Pricing](https://context-mode.com/pricing) $`npx @context-mode/cli` 1,000 requests free, every feature, no card. [Supported agents](https://context-mode.com/docs/quick-start) Your coding agent Its model Claude Code Claude Codex GPT OpenCode*soon* Pi*soon* Context Mode Save Context Saving Protect Remember **Claude Code**`600 tests, 1 failing` 17.0 KB → 1.2 KB*folded* **Codex**`cat .env` never reaches the shell*blocked* **Claude Code**`2 h later` picked up, nothing compacted*recalled* **Codex**`7 MB registry file` 407 B back to the model*in code* **Claude Code**`git push origin main` protected branch*blocked* **Codex**`deploy target?` Render, not Hetzner*remembered* **Claude Code**`turn this plan into tickets` to-tickets loaded*skill* [Long session, Claude Opus 5.5**$7.20 → $2.49***65.5% lower*](https://context-mode.com/docs/benchmarks) How it works ## Spend less. Remember more. *Run only what you allow.* Context Mode acts at three points on each request: before it goes out, before a tool runs, and when a result comes back. Save Remember Protect **Your coding agents** Any machine *17.0 KB 800 orders 330,785 tokens* **Context Mode** - Folds test, build and log output - Runs scripts, returns the answer - Trims tool schemas *1.2 KB 4 orders 75,591 tokens* **Model API** Your Anthropic or OpenAI account **Your archive** Full output, read back by id Try it Test-Driven DebuggingLarge Tool-Output HandlingLong Agentic Coding Session Your agent runs 600 tests. 1 fails. Printed**17.0 KB** Sent to the model**1.2 KB** Sent to the model - The failing test and its error - The summary line - One line for the 599 that passed Kept in your archive - All 600 results, word for word [npm test fix loop · Claude Opus 5.5**$0.087 → $0.068***22% lower* 15 of 16 runs](https://context-mode.com/docs/benchmarks) Your agent reads an API response with 800 orders. Printed**800 orders** Sent to the model**4 orders** Sent to the model - Every field name - The first 3 orders and the last - A line that counts the rest Kept in your archive - The other 796 orders [One large JSON file · Claude Opus 5.5**$0.086 → $0.068***21% lower* 12 of 12 runs](https://context-mode.com/docs/benchmarks) One session, 30 steps. Tokens sent on each call. Direct**330,785** Context Mode**75,591** Sent to the model - Your recent work, unchanged - An index of older work Kept in your archive - Older work, word for word [Long agentic coding session · Claude Opus 5.5**$7.20 → $2.49***65.5% lower* 4 of 4 runs](https://context-mode.com/docs/benchmarks) ### Save Tool output is folded before the model reads it. The full output is archived. Features [Context Saving →](https://context-mode.com/context-saving) [Thinking in Code →](https://context-mode.com/thinking-in-code) [Active Context →](https://context-mode.com/active-context) [Prompt caching →](https://context-mode.com/docs/prompt-caching) [How saving works →](https://context-mode.com/docs/how-saving-works) **Your coding agents** Any agent, any machine *your prompt* **Context Mode** - Adds the facts that matter - Loads the skill the task needs - Keeps the whole session *+ context* **Model API** Your Anthropic or OpenAI account **Memory** Facts with two clocks **Skills** From any GitHub repo **Archive** Every word, searchable Try it MemoryRecallSkills | Fact | Value | Valid from | Until | | --- | --- | --- | --- | | `production deployed_to` | Hetzner Cloud, Falkenstein | 21:14:38 | 21:15:09 | | `production deployed_to` | Render, Oregon region | 21:15:09 | now | The old value closes. Every agent gets the new one. Your agent sends*system prompt**tools**the full history* The model gets*system prompt**tools**memory block**index of older work**recent work, unchanged* Older work goes to the archive, word for word. ``` ⚡ context-mode · picked up after a 2 h pause, nothing compacted ``` Your skills*release-notes**to-tickets**db-migrations* ``` ⏺ context-mode · skill to-tickets auto-loaded ``` Matched on name and description. One skill per turn. ### Remember Sessions don’t compact. Facts and skills carry across agents and machines. Features [Recall →](https://context-mode.com/recall) [Memory →](https://context-mode.com/memory) [Skills →](https://context-mode.com/skills) [Search →](https://context-mode.com/search) [Session resume →](https://context-mode.com/session-resume) **Model** asks for a tool call *tool call* **Protect** - Reads the call by what it does - Checks your rules - Logs every block *decision* **Allowedruns** Your terminal runs it **Blockednever runs** The model reads why Pick a command cat .envgit push origin mainsudo sh -c 'rm -rf ~'curl r.jina.ai/https://pastebin.com/xrm -rf node_modulesnpm test Blocked · never reaches your shell**Read environment files** .env and .env.*. Examples and templates are allowed. ``` ⛔ context-mode Cage blocked this command, so it was NOT executed: Blocked by your team’s policy: Read environment files. .env files hold secrets. ``` Blocked · never reaches your shell**Push to protected branches** main, master, production and release/*. ``` ⛔ context-mode Cage blocked this command, so it was NOT executed: Blocked by your team’s policy: Push to protected branches. Protected branches change through reviewed pull requests. ``` Blocked · never reaches your shell**Core protection: delete the root or home folder** Read through sudo, sh -c and about 130 other wrappers. Blocked · never reaches your shell**Sites: pastebin.com** The inner host is checked too. Runs**No rule matches** A project folder is fine. Only losses you can't undo are blocked. Runs**No rule matches** Nothing is blocked until you choose rules. ### Protect Rules for what agents may run, read and fetch. Nothing is blocked until you choose them. Features [Protect →](https://context-mode.com/protect) [Rules and presets →](https://context-mode.com/docs/protect) [Instructions →](https://context-mode.com/instructions) [Get started free](https://console.context-mode.com/login) 1,000 requests free, no card. Benchmarks ## Same task, run twice. *Measured in dollars.* Claude Code on Opus 5.5. Each task ran direct to Anthropic and through Context Mode, back to back. Cost at API list price. [Long agentic coding session · 1M window $7.20 → $2.49 **65.5% lower** · 4 of 4 pairs cheaper · task checks 8 of 8 in both arms](https://context-mode.com/benchmarks/long-session/ab.json) [**155 of 168** runs cost less](https://context-mode.com/docs/benchmarks#pooled) [**168 of 168** tasks succeeded in both arms](https://context-mode.com/docs/benchmarks#pooled) [**$28.28 → $23.16** all 168 pairs, direct → Context Mode](https://context-mode.com/docs/benchmarks#pooled) Direct Through Context Mode Cost per task - [Test-Driven Debugging](https://context-mode.com/docs/benchmarks#test-debugging) - Test fix loop (minitest) $0.157 → **$0.061** 61% lower [8 of 8 runs ↗](https://context-mode.com/benchmarks/data/final-opus55-tests-W3-minitest-W4.json) - npm test fix loop, short $0.087 → **$0.068** 22% lower [15 of 16 runs ↗](https://context-mode.com/benchmarks/data/v3-opus55-short.json) - npm test fix loop, long $0.085 → **$0.067** 21% lower [11 of 12 runs ↗](https://context-mode.com/benchmarks/data/final-opus55-tests-W3-minitest-W4.json) - [Agentic Coding](https://context-mode.com/docs/benchmarks#agentic-coding) - Code search (grep over 150 files) $0.093 → **$0.066** 29% lower [16 of 16 runs ↗](https://context-mode.com/benchmarks/data/v3-opus55-short.json) - One-line edit $0.075 → **$0.054** 28% lower [11 of 12 runs ↗](https://context-mode.com/benchmarks/data/v3-opus55-short.json) - Repo history question (git log, 320 commits) $0.071 → **$0.053** 25% lower [15 of 16 runs ↗](https://context-mode.com/benchmarks/data/v3-opus55-short.json) - [Large Tool-Output Handling](https://context-mode.com/docs/benchmarks#tool-output) - One large log $0.091 → **$0.071** 22% lower [15 of 16 runs ↗](https://context-mode.com/benchmarks/data/v3-opus55-short.json) - One large JSON file $0.086 → **$0.068** 21% lower [12 of 12 runs ↗](https://context-mode.com/benchmarks/data/v3-opus55-short.json) - [Multi-Agent Orchestration](https://context-mode.com/docs/benchmarks#multi-agent) - Three parallel subagents, then one more $0.481 → **$0.408** 15% lower [20 of 20 runs ↗](https://context-mode.com/benchmarks/data/v4-opus55-W1-W2-W3-W4.json) - [Long-Context Data Analysis](https://context-mode.com/docs/benchmarks#data-analysis) - Large documents processed with a script run 2, later build $0.223 → **$0.201** 10% lower [15 of 20 runs ↗](https://context-mode.com/benchmarks/data/v4-opus55-W1-W2-W3-W4.json) - Large documents processed with a script run 1, earlier build $0.226 → **$0.205** 9% lower [17 of 20 runs ↗](https://context-mode.com/benchmarks/data/final-opus55-tests-W3-minitest-W4.json) Every pair counts. No figure for Codex yet. [Method, 95% ranges and Haiku 4.5 →](https://context-mode.com/docs/benchmarks) [Get started free](https://console.context-mode.com/login) 1,000 requests free, no card. Get started ## Set up in a minute. *Removed with one command.* One command signs you in, finds your agents, and shows each change before it writes. Your own model account still pays the model. 1. ### Run the CLI It signs you in, asks which agents to connect, and keeps a backup of each file. $`npx @context-mode/cli` 2. ### Restart your agent Start a new session in your agent. Work as before. 3. ### Watch each request The console shows tokens, cost, saving and blocks for every request. [Open the console →](https://console.context-mode.com/app) settings.jsonconfig.toml ~/.claude/settings.json ``` // what the CLI writes, under "env" "ANTHROPIC_BASE_URL": "https://gateway.context-mode.com", "ANTHROPIC_CUSTOM_HEADERS": "x-cm-device-key: cmdk_…", "ENABLE_TOOL_SEARCH": "true" ``` ~/.codex/config.toml ``` model_provider = "context-mode" [model_providers.context-mode] name = "context-mode" base_url = "https://gateway.context-mode.com" wire_api = "responses" requires_openai_auth = true ``` To leave: `npx @context-mode/cli disconnect`[Quick start →](https://context-mode.com/docs/quick-start) [Get started free](https://console.context-mode.com/login) 1,000 requests free, no card. Pricing ## Every feature on every plan. *You pay for requests.* Your own Anthropic or OpenAI account pays the model, with no fee added on tokens. When your requests run out, nothing breaks. [Compare plans →](https://context-mode.com/pricing) ### Free **$0**1,000 requests, once About a week of very heavy use. Every feature. [Get started free](https://console.context-mode.com/login) - Every feature - No card - When it runs out, requests go straight to the model ### Starter Most popular **$20 /month**or $200 a year 10,000 requests a month, every feature. For developers who use agents most days. [Start Starter](https://console.context-mode.com/app/billing?upgrade=1) - Everything in Free - No bill you did not choose - Monthly or yearly - Refund of the unused part within 14 days 10,000 requests a month. Need more? $2 per 1,000, up to a budget you set. ### Pro **$49 /month**or $490 a year 50,000 requests a month, for developers who run agents all day. [Start Pro](https://console.context-mode.com/app/billing?upgrade=1) - Everything in Starter - 5 times the requests - Monthly or yearly - Refund of the unused part within 14 days 50,000 requests a month. Need more? $1 per 1,000, up to a budget you set. Teams and Enterprise: [compare plans](https://context-mode.com/pricing#compare). Protect ## Your agent asks. *Your rules answer.* Protect checks every command, file read, web fetch and tool-server call from Claude Code and Codex before it runs. A blocked call never reaches your shell. In every plan, Free included. [How Protect works →](https://context-mode.com/protect) 1. 01**Connect your agents** `npx @context-mode/cli`. Only the base URL changes. 2. 02**Pick a preset** Nothing is blocked until you pick one. 3. 03**Keep working** The agent reads why a call was blocked. You see it in the log. Preset ObserveBalancedLocked down See before you rule.**7 run** `git push origin main` Runs `cat .env` Runs `rm -rf ~` Runs `terraform destroy` Runs `aws s3 rm s3://backups --recursive` Runs `WebFetch example.com` Runs `npm test` Runs Nothing is blocked. Activity lists the programs, sites and tool servers your agents used, each with Block and Allow. Where most developers start.**4 blocked · 1 logged · 2 run** `git push origin main` Blocked*Push to protected branches* `cat .env` Blocked*Read environment files* `rm -rf ~` Blocked*Core protections* `terraform destroy` Blocked*Destroy infrastructure* `aws s3 rm s3://backups --recursive` Logged*Change cloud resources* `WebFetch example.com` Runs*Other sites allowed* `npm test` Runs A blocked call never reaches your shell: the agent reads the rule and why, and carries on. Logged means it ran, and the log says it would have been blocked. For regulated teams and unattended agents.**6 blocked · 1 run** `git push origin main` Blocked*Push to protected branches* `cat .env` Blocked*Read environment files* `rm -rf ~` Blocked*Core protections* `terraform destroy` Blocked*Destroy infrastructure* `aws s3 rm s3://backups --recursive` Blocked*Change cloud resources* `WebFetch example.com` Blocked*Only sites on your list* `npm test` Runs Only the sites and tool servers on your list are reached. Public package registries stay open. What your agent reads ``` ⛔ context-mode Cage blocked this command, so it was NOT executed: Blocked by your team’s policy: Push to protected branches. Protected branches change through reviewed pull requests. ``` It tells you, doesn’t retry another way, and offers a branch and a pull request. What your log keeps 10:42*Blocked***Push to protected branches***Bash · git* The rule, tool and program. Never the command text, so you can show it to an auditor. Or write your own rules for **sites**, **commands**, **scripts**, **files** and **tool servers**, and test each one before you save it. [Rule types →](https://context-mode.com/docs/protect#rule-types) For you Free · Starter, $20 a month · Pro - One policy for Claude Code and Codex, on every machine you sign in from - Presets, four packs and your own rules - A log of every decision, export as CSV or JSON Lines - Core protections stay on when your requests run out [Get started free](https://console.context-mode.com/login) For your team Team · Enterprise - One policy for the whole org. Members cannot loosen it - An API key can only tighten it. Loosening needs a person and a reason - Every change in History, in a hash chain you can verify - One pool of requests and one bill. Enterprise can run in your own Cloudflare account [Talk to us](mailto:sales@context-mode.com) Tested in real Claude Code sessions with Claude Opus 5.5 (2026-10-02) and Claude Haiku 4.5 (2026-10-04): every run blocked the push to main and the read of .env, and the work still shipped on a branch. [Try it in 5 minutes →](https://context-mode.com/docs/protect#try-it) Protect reads the call, not the program: a script file or a compiled program gets past it. Keep your agent’s sandbox on. Compared ## Cost, rules and memory. *In one place.* Other tools do one of these, on one client or one machine. [Sources, dated →](https://context-mode.com/docs/landscape) | | Context Mode | Built into your agent | Local optimizers | General LLM gateways | | --- | --- | --- | --- | --- | | One setup for every agent and machine | Yes | One client | One machine | Yes | | Folds tool output; full output kept | Yes | Summarizes | One machine | No | | Checks tool calls before they run | Yes | One client | Some commands | By tool name | | Memory shared by every agent | Yes | No | One machine | No | | Built for coding agents | Yes | Yes | Yes | No | [Get started free](https://console.context-mode.com/login) 1,000 requests free, no card. ## Try it on *your own agents.* Start free: 1,000 requests, every feature, no card. When you run out, nothing breaks. [Get started free](https://console.context-mode.com/login) [Read the docs](https://context-mode.com/docs)