Provider packages translate a shared agent loop into model-specific request, cache, and session behavior.
earendil-works/pi
A compact, extensible coding-agent stack whose package boundaries and session model make provider behavior, compaction, rendering, and customization legible.
Developers who want a programmable terminal agent and reusable agent packages.
terminal coding agent / agent packages / extensions
Session affinity is evidence, not one reusable header
Provider-specific session identifiers and expiry behavior are kept in typed session state instead of being flattened into one generic transport assumption.
System Map
Eight inspection boundariesConversation entries, compaction records, and provider session metadata remain distinct forms of state.
The agent package exposes tools and extensions while the terminal client owns interaction and rendering.
Tool availability and host authority remain deployment choices that extensions must make visible.
Durable session entries, compaction records, and affinity metadata preserve resumable work.
Delegation is not the primary teaching surface; package composition and session continuity are better-evidenced boundaries.
The terminal renders durable evidence, cache pressure, and session state for operator inspection.
Typed entries, rendered session artifacts, tests, and package boundaries provide the strongest receipts.
Learn Through This Project
8 lessonsLatest Reporting
8 articlesOctober 18, 2026 — Pi’s experimental durable harness now stores a sleeping task’s deadline and next step, letting a host alarm bring execution back later. The change separates continuing work from a continuously running process. earendil-works/pi
Pi and Crush Tell the Terminal When They Need YouOctober 9, 2026 — Two released coding agents now share a status protocol for working, waiting and finished runs. It gives compatible terminals a way to surface the agent that needs attention—without deciphering each tool’s screen. charmbracelet/crush, earendil-works/pi, charmbracelet/bubbletea, openai/codex
Pi Adds Screenshot Questions to Its Classifier InterfaceOctober 8, 2026 — Pi 1.1.0 lets an agent script send images to OpenAI Decisions and receive answers it can use in code. The release expands a practical division of labor: the main model writes the task, a separate call judges a bounded question, and the script chooses what happens next. earendil-works/pi
Pi Adds Built-In MCP—and Keeps the Tools ReplaceableSeptember 30, 2026 — The coding agent once left MCP to outside extensions. Its new release supplies an implementation, while giving extension authors a shared way to call tools and the ability to replace the defaults. earendil-works/pi
Inside TypeSafe’s Model Comparison: What the Adapter Makes You Pay ForSeptember 29, 2026 — A probability-bearing answer buys more than a label. Local decision models and a public routing experiment now expose the next choice: how much accuracy must the cheaper route preserve? typesafe-ai/system-one-adapter-python, unslothai/unsloth, earendil-works/pi, ickma2311/jev-baselines-eval
API Discovery: Shorten the Context, Keep the HistorySeptember 22, 2026 — Pi v0.87.0 lets a builder append an edit to future model context while leaving the original transcript intact. That is useful for a reading assistant—and it is not data deletion. earendil-works/pi
Pi's New Harness Writes the Restart Point Before It ActsAugust 20, 2026 — Pi's dev branch is turning an agent run into a durable, remotely attachable session—but its most important promise is also the simplest: record the restart point before the model or tool gets to act. earendil-works/pi
API Discovery: Pi’s xAI Route Keeps Reasoning State in PlayAugust 17, 2026 — Pi now sends every built-in xAI model through the Responses API—a useful route for agents that need a deliberate, client-owned continuity plan rather than a model-name swap. earendil-works/pi
Primary Sources
Open Questions
- Which host permissions and extension effects can a shared session prove after the fact?
Evidence state describes what this Atlas profile can support today. It is not a product score and it is not evidence that uninspected capabilities are absent.