pi-openai-long-context

Raise GPT-5.6 and GPT-6 models (including GPT-6 Astra) from pi's 272K context window to 1.05M, one task at a time.

Install

pi install npm:pi-openai-long-context

Restart pi, or run /reload.

Use it

/long-context

Toggles the big window on and off for the supported model you are using. A ⚠ in the footer means it is on. The command only appears in the / menu for gpt-5.6-* and gpt-6-* models on openai or openai-codex.

It turns itself off

In ordinary sessions, switching models, starting a new session, restarting pi, /reload — all of it drops you back to 272K, so you cannot leave it on and be billed for it later. Nothing is saved to your settings.

Compaction is the exception: it keeps the big window, which is the point of turning it on.

If turning it off would immediately trigger automatic compaction, pi asks whether to compact or keep long context instead.

pi-subagents

Automatic long context for pi-subagents is off by default. To opt in once for child sessions in its background runner processes, create ~/.pi/agent/openai-long-context.json:

{
  "autoEnableSubagents": true
}

If you use PI_CODING_AGENT_DIR, put the file in that directory instead. Only the boolean true enables this behavior; missing, false, unreadable, or invalid config leaves it off. This extension never creates or changes the config file.

After opting in, child sessions in a marked runner automatically enable long context at startup when this extension is loaded and the selected model is supported. No per-launch flag, /long-context command, or extensionBindings is needed, and the main session's toggle is unchanged. /long-context remains available without opting in.

The PI_SUBAGENT_CHILD=1 marker identifies the runner process, not an individual child's execution mode. The opt-in therefore also covers nested foreground children (async: false) in that process when they load this extension. Foreground children outside a marked runner remain manual.

Background children normally discover installed extensions; foreground children need an explicit extension path. If necessary, include this extension's index.ts path in the agent's extensions or subagentOnlyExtensions configuration. Extension-denying policies still apply.

Each child keeps its own context window. Compaction preserves it; toggling off, switching models, or shutting down restores the previous value. Starting or reloading a marked child session automatically enables it again only while opted in. To opt out, set autoEnableSubagents to false or remove the file; already-running children are unaffected. The long-context costs below apply to opted-in children too.

What it costs

Past 272K input tokens, OpenAI bills the whole request at its long-context rate — see OpenAI's pricing. On a subscription, that also burns through your quota faster.

Supported models and providers

Works on gpt-5.6-* and gpt-6-* models on the openai and openai-codex providers. Other providers are left untouched.

The toggle uses 1.05M tokens, the documented maximum for GPT-5.6 and GPT-6 Astra. Future gpt-6-* models are matched automatically, but this extension does not validate their limits; verify each model's documented context window before enabling it.

If you already set your own context window for a supported model in models.json, turning this off restores your value, not pi's.

License

MIT