ranxianglei/billion-context
The official billion-context plugin: a context-compression plugin for small context windows (a 100K context is enough), token savings (5x fewer tokens), and month-long single sessions (billions of tokens).
Install
via npm
Resolves a published tarball rather than cloning the repository, and installs without any extra setup. Swap `web` for your profile name if you run another one.
via GitHub · npm package
Installing from GitHub runs the project's build script, which pnpm blocks until you allowlist it — run the command once and pnpm prints the exact key to add under `allowBuilds` in ~/.dsh/profiles/web/pnpm-workspace.yaml.
The desktop app keeps its own profile, desktop, and only the dsh command it ships may change it — the npm-installed dsh refuses that profile. Install the command from the app menu (Manage dsh Command…), open Desktop once, quit it fully, run the line below, then reopen Desktop.
via npm
Resolves a published tarball rather than cloning the repository.
via GitHub · npm package
Installing from GitHub runs the project's build script, which pnpm blocks until you allowlist it — run the command once and pnpm prints the exact key to add under `allowBuilds` in ~/.dsh/profiles/desktop/pnpm-workspace.yaml.
Our install check ran in a web profile on npm's dsh, not in the desktop app.
What happened when we ran it
Installed cleanly when we ran it
Every command here is run in a throwaway container against a clean profile, and the result is whatever the harness recorded — not a guess from the source. Last run 6d agoon DeepSeek Harness 0.2.1-rc.5.
Show it in your README
For maintainers: the badge serves this listing's latest sandbox verdict, so a re-run updates it on its own — and it links readers to the full result here.
Due diligence
Security
No build script was blocked in our run.
Source: our sandbox install on DeepSeek Harness 0.2.1-rc.5, 6d ago.
package.json declares no preinstall, install or postinstall script.
Source: package.json on GitHub, read at our last sync. prepare is not recorded; it runs only on a GitHub install, and shows up under build scripts when pnpm blocks it.
shell commands (bash, exec, spawn …)
Source: a keyword match on the README — a hint about what the plugin touches, not an analysis of its code.
Not an audit. These are the record of one install and two text scans; nobody has read this plugin's code on your behalf. Plugins run with your agent's permissions.
What it is — A DeepSeek Harness plugin that shrinks a coding agent's session context so long stretches of work can stay inside a small window.
Who it is for — You're on a model with about 100K context, pushing one coding task across days, and the transcript keeps filling the window until you have to start over. Frequent model-changers can pass: the README lists a model switch as a usual reason prefix-cache hit rate falls, which can erase the token savings.
Watch out — The sandbox only confirmed a clean install on a fresh profile and registration by DSH. The plugin runs shell commands. The 5x token cut, month-long sessions, and 95–97% prefix-cache hit rate are README figures and were not measured here.
The verdict — I would try it on a throwaway profile and read /acp before letting it touch a session I already care about.
Generated by grok-4.6, and a starting point rather than a verdict. Where it says a plugin installs or does not, that is from a real run in a clean profile — everything else is read off the repository. Trust the source over this.
billion-context is a DeepSeek Harness plugin that compresses long agent conversations so they can continue within a smaller model context window. DeepSeek Harness loads billion-context between an agent and its model API: it rewrites Anthropic and OpenAI streams through acp-kernel, while the model decides when to compress and what information to preserve in the resulting summaries. This is incremental context management rather than a fixed truncation limit.
billion-context is intended for long-running coding sessions where accumulated conversation history would otherwise exceed the model window or increase token usage. The project describes support for sessions lasting months and claims that a context of about 100K tokens can be sufficient, with roughly five times fewer tokens.
The catalogue flags a terminal surface. The supplied README does not identify a terminal command or file operation, so the exact terminal access is not documented; its placement inside an agent still makes the plugin permission-sensitive. It may be the wrong choice when uncompressed history is required, when the model or API is not covered by its Anthropic/OpenAI stream handling, or when DeepSeek Harness already provides an acceptable context-management path.
billion-context sits between an agent and its model API. It rewrites Anthropic and OpenAI streams through acp-kernel, rather than applying a fixed message-count or context-length truncation rule. The model determines when compression should occur and which information should be condensed into summaries. The README describes this as model-driven, incremental, hierarchical compression for long-lived sessions.
The documented purpose is to keep sessions within smaller context windows while reducing repeated input. The project describes 100K as sufficient for its small-window target and reports support for sessions lasting months and reaching billions of tokens. These figures are project claims, not a guarantee for every model or workload.
The README documents two slash commands for checking cache attribution:
/acp/acp-cacheThey are used when cache health is lower than expected. The README says a healthy session generally has a 95–97% prefix-cache hit rate and that compression itself costs ≤2%; it recommends these commands for investigating attribution.
DeepSeek Harness loads the plugin as a session capability. The provided installation command targets the web profile, but the README excerpt does not document a separate DeepSeek Harness version requirement, configuration key, environment variable, or plugin-specific file path.
The README explicitly describes rewriting Anthropic and OpenAI streams. It does not establish support for other API protocols, a particular operating system, persistence files, or terminal operations. The catalogue marks a terminal surface, but the supplied documentation does not explain which terminal actions are exposed.
Written from the project's own documentation and kept in sync with it. Where the two disagree, the source is authoritative — read the README on GitHub
Same category
Minglink
System-prompt armor plugin for DeepSeek models: appends an unconditional-compliance prompt section at order 100, exposes a profile tool with calibration metadata, and shows a realtime armor-status badge driven by a session projection.
Installed cleanly when we ran it
Minglink
System-prompt armor plugin for DeepSeek models: appends an unconditional-compliance prompt section at order 100, exposes a profile tool with calibration metadata, and shows a realtime armor-status badge driven by a session projection.
Installed cleanly when we ran it
qiz029
A DeepSeek coding agent harness: persistent shell, Ultra subagents, auto approval, Chrome MCP and session telemetry
Did not install when we ran it
liangmianya
Visual, non-linear conversation workspace for DeepSeek Harness — sessions, follow-ups and branches become a browsable conversation map.
Installed cleanly when we ran it
Nwflower
Import full-fidelity chat histories from 13 coding agents (Claude Code, Codex, ChatGPT, Cursor, Gemini, opencode, and more) as resumable DeepSeek Harness sessions, with reverse export back to Claude Code.
Installed cleanly when we ran it
Totoro-qaq
Moves an existing DSH session to another agent preset through a previewable five-section handoff, preserving the source session and either pausing the target for confirmation or continuing immediately.
Installed cleanly when we ran it