Context-savers

A context-compression plugin — small context windows (a 100K context is enough), 5x fewer tokens, month-long single sessions (billions of tokens), and compression quality — for all agents: pi, OpenCode, Codex, Claude Code, and more. billion-context is all

MCP plugin that saves 98% of your context window. Works with Claude Code, Gemini CLI, VS Code Copilot, OpenCode, and Codex CLI. Sandboxed code execution, FTS5 knowledge base, and intent-driven search.

Subagent orchestration for the pi coding agent — a thin dispatcher runs parallel subagents under a verification contract, keeping their output out of its context window. Multi-agent fleets, 15 personas, skills, herdr panes, peer coms, Hermes desktop.

Codex-style context windows for Pi: durable reset windows, session history tools, and persistent notes.

Pi extension that keeps noisy tool output out of your context window. Automatically rewrites shell commands through Hypa for local, deterministic compression, context-aware file tools, and recoverable evidence.

Multi-phase security audits with specialist sub-agents, isolated context windows, capped concurrency, and resumable state — packaged as a Pi extension.

pi coding agent extension: additional auto-compact thresholds (percent of context window / used tokens), taking the lowest of all thresholds including pi's built-in remaining-tokens threshold

Toggle OpenAI GPT-5.6 and GPT-6 models between pi's 272K context window and 1.05M tokens.

Branch-aware mid-context compression for the Pi coding agent

Pi extension: detects and interrupts infinite thinking-block and tool-call loops in real time before they exhaust your context window.

Rich custom status bar for pi with token usage, cost, DeepSeek balance, context window, and timing.

Extension for the pi coding agent: enforce a context token budget (defaulting to the model's configured context window; 200k when unknown) by forcing compaction, including mid-turn during long tool loops, instead of riding long-context models up to their

Pi extension that shows context window usage visualization

Posthorse: fresh context, same journey. Native no-summary context windows for the fitchmultz/pi fork, with sparse budget reminders, handoffs, notes, and history.

Delegate tasks to specialized subagents with isolated context windows.

Custom status-bar footer for the pi coding agent — model, thinking level, path, git branch, token usage, cost, context window usage, and cache hit rate, and generation throughput.

Proactive auto-compaction: when the context window crosses a threshold, compact between turns so a long session never hits the wall or needs a restart

Session-native Notes, History, and agent-owned Context Windows for Pi.

为 Pi Agent 打造的输出精简提示插件 — 多档力度压缩输出 token 省钱 | A prompt-verbosity plugin for Pi Agent — tiered output compression to save tokens & cost

Headroom context compression for pi with a working headroom_retrieve tool

Pi Coding Agent extension for OmniRoute — view combos, browse providers, and sync models with enriched metadata (context windows, max tokens, reasoning, and vision) to the Ctrl+P picker

Intelligent context window management for pi: bash output summarization, silent meta-amnesia pruning, lazy skill/tool loading, cache-safe tail pruning, and per-turn token accounting.

Subagent tool for pi: delegate tasks to specialized agents with isolated context windows (single, parallel, chain modes) with streaming output, plan-mode propagation and parent-model inheritance

Pi extension that provides logging of the current context window with each llm call as well as ability to remove thinking/reasoning text from the context window.

Active Context Pruning (ACP) extension for Pi — model-driven context compression that keeps conversations flowing.

A pi extension that caps effective model context windows at 200k tokens for earlier auto-compaction.

Transparent LLM context compression for Pi using Headroom

Pi extension that guards context window growth by auto-limiting read and rg output

In-process shared-prefix concurrent fork/join for the Pi coding agent. Read-only forks by default; write forks are isolated in their own git worktrees. The parent keeps its context window; children return dense structured reports.

Log in to the airpx LLM proxy (https://airpx.cc) with an sk-proxy API key and auto-import its model catalog (prices, context windows, reasoning, vision) into the pi coding agent.

A sliding context window for Pi coding agents. When a chat gets long, old messages are saved to a small local database instead of being dropped — so your agent can search them and pull the exact one back whenever it needs it.

Tag Pi in when Claude Code or Codex hits rate limits, context windows, or crashes. Seamlessly transfer context, uncommitted git work, and subagent tasks.

MCP plugin that saves 98% of your context window. Works with Claude Code, Gemini CLI, VS Code Copilot, OpenCode, and Codex CLI. Sandboxed code execution, FTS5 knowledge base, and intent-driven search.

Per-model context-usage thresholds for Pi's built-in auto-compaction, so models with different context windows and performance profiles compact at the right time

OpenAI toolkit for Pi: Codex Remote Context windows, remote compaction v2, routed Web Search, image gen, auto mode.

Pi extension that automatically compacts large tool outputs before they enter the model context window, saving tokens while preserving full output as recoverable local artifacts.

Auto-discover local model providers (Ollama) and register their models with Pi on startup. Per-model context windows and capabilities, no /reload required.

Pricing-aware model picker for pi: select models with per-1M-token cost, context window, cache rates, tiers and reasoning support. Sort models by input/output/total cost with ctrl+s.

pstack for Pi. Poteto's rigorous engineering workflow ported to the Pi Agent Harness: 50 skills and 2 agents, plus an extension that wires the mode into Pi settings, sessions, compaction, and the context window.

Token reduction & persistent memory for Pi coding agent — compress tool outputs, save decisions across sessions, survive compaction, backed by Obsidian

MCP plugin that saves 98% of your context window. Works with Claude Code, Gemini CLI, VS Code Copilot, OpenCode, and Codex CLI. Sandboxed code execution, FTS5 knowledge base, and intent-driven search.

Ollama (local + Ollama Cloud) and LM Studio providers for the pi coding agent - live model discovery with real capabilities (tools/vision/thinking) and context windows.

Filter and block models from any provider via declarative allow/block rules. Hide unwanted Copilot models, enforce global allowlists, or block by context window size.