Posthorse ships as a native, no-summary context window system for the fitchmultz/pi fork of the Pi coding agent, letting models continue work across fresh context windows without summaries. It includes rollover tools, durable notes, and history recovery so a full transcript remains recoverable while the model gets clean context.
The maintainer of `resume-from` released a local session handoff tool that lets developers continue coding sessions across Pi, Claude Code, and Codex without sharing raw vendor-specific session files. It addresses the problem of rebuilding task context when switching agents or hitting usage limits.
The maintainer of `resume-from` released a local session handoff tool that lets developers continue coding sessions across Pi, Claude Code, and Codex without sharing raw vendor-specific session files. It addresses the problem of rebuilding task context when switching agents or hitting usage limits.
A GitHub project implements Zero-Mem, a zero-token long-term memory system for the pi coding agent, based on arXiv:2607.29377. It stores conversation traces in an entity-context graph and temporal hierarchy, retrieving relevant evidence without extra LLM calls or tokens.
A GitHub project implements Zero-Mem, a zero-token long-term memory system for the pi coding agent, based on arXiv:2607.29377. It stores conversation traces in an entity-context graph and temporal hierarchy, retrieving relevant evidence without extra LLM calls or tokens.
A developer building long-running analytics agent workflows reports that durable agent memory for semantic continuity now works, exposing a second-order problem: managing active context during a session. They frame context as separate layers — stable instructions, durable project context, and intermediate exploration/tool output — and ask how others handle the transient layer. The post matters to practitioners building stateful agents beyond single-session use.
A developer building long-running analytics agent workflows reports that durable agent memory for semantic continuity now works, exposing a second-order problem: managing active context during a session. They frame context as separate layers — stable instructions, durable project context, and intermediate exploration/tool output — and ask how others handle the transient layer. The post matters to practitioners building stateful agents beyond single-session use.
pisesh is a terminal UI for managing Pi coding-agent sessions, helping users who run Pi from multiple project folders avoid opening the wrong session. It provides session list favorites, text search, custom names, a current-project filter, and optional title generation. It is a single Node script with no runtime dependencies, and version 0.3.0 improved session handoff by using Pi's own session-switch API.
pisesh is a terminal UI for managing Pi coding-agent sessions, helping users who run Pi from multiple project folders avoid opening the wrong session. It provides session list favorites, text search, custom names, a current-project filter, and optional title generation. It is a single Node script with no runtime dependencies, and version 0.3.0 improved session handoff by using Pi's own session-switch API.
Pi-CWL is an agent framework that runs a single session through over 80 million tokens without needing context compaction, avoiding information loss and hallucination. It solves the finite context window problem for long-running agent tasks, targeting developers building complex multi-step agent workflows.
Pi-CWL is an agent framework that runs a single session through over 80 million tokens without needing context compaction, avoiding information loss and hallucination. It solves the finite context window problem for long-running agent tasks, targeting developers building complex multi-step agent workflows.