aidev TIER 1 harness Mario Zechner / earendil-works

Pi

RELEASES30
BREAKING5
LATESTv0.84.2
SHIPPED11d
TRACKED SINCE2026-06-04

RELEASE CADENCE · LAST 12 MONTHS

09
10
11
12
01
02
03
04
05
06
07
08
The numbers behind the bars
Releases per month, last twelve months
MonthReleasesShare
2025-09 0 0%
2025-10 0 0%
2025-11 0 0%
2025-12 0 0%
2026-01 0 0%
2026-02 0 0%
2026-03 0 0%
2026-04 0 0%
2026-05 0 0%
2026-06 16 53%
2026-07 11 37%
2026-08 3 10%
12-MONTH TOTAL 30—

BREAKING CHANGES

v0.84.0 Pi 0.84.0: Fullscreen TUI, Mermaid/LaTeX rendering, per-directory context overrides, extensive API breaks 19d v0.83.0 Pi v0.83.0: credential export, headless OpenRouter login, Claude Opus 5, TypeBox breaking changes 27d v0.80.8 Pi v0.80.8: ModelRuntime, xAI OAuth, Grok 4.5, SDK breaking changes 1mo

RECENT

Density
11d
10:14
aidev pi

Pi v0.84.2: fullscreen transcript search, configurable default tools, strict JSON sampling, Cloudflare AI Gateway binding

Fullscreen mode gains transcript search (Ctrl+Shift+F) with incremental highlighting and navigation. Default built-in tools are now configurable globally or per project. Experimental strict JSON-schema constrained sampling added for default tools. Cloudflare Workers AI Gateway routing through binding without API token. Mistral SDK replaced with native HTTP transport. Multiple TUI, extension, and subagent fixes.

featurefixapi-change
19d
06:07
pi Pi 0.84.1: Qwen Token Plan provider, auth check, fullscreen selection v0.84.1 · SIG 2/5
19d
11:07
aidev pi BREAKING

Pi 0.84.0: Fullscreen TUI, Mermaid/LaTeX rendering, per-directory context overrides, extensive API breaks

Major release introducing fullscreen TUI mode with independently scrollable transcript and draggable scrollbars, Mermaid diagram and LaTeX math rendering in interactive transcripts, per-directory context overrides via AGENTS.override.md, advanced custom model sampling parameters, and a built-in Baseten provider. Includes extensive breaking changes to the provider API (ModelsStreamTransforms→ModelsRequestTransforms, message_update event deltas, ModelRegistry/provider refresh APIs, v4 lane-based session model replacing the legacy harness).

featurebreakingapi-changecapability
27d
22:30
aidev pi BREAKING

Pi v0.83.0: credential export, headless OpenRouter login, Claude Opus 5, TypeBox breaking changes

Pi v0.83.0 adds credential export commands for external clients, headless OpenRouter sign-in over SSH, and Claude Opus 5 support via GitHub Copilot. The bundled TypeBox aliases were upgraded to 1.3.7 with several removed APIs, a breaking change for extensions using deprecated features. Dozens of fixes include session streaming, context file loading, Git package installs, and llama.cpp token accounting.

featurebreakingmodel-supportfix
1mo
12:47
aidev pi

Pi v0.82.1: Claude Opus 5 support, Anthropic bearer auth, faster model catalogs

Claude Opus 5 is now available on Pi through Anthropic and Amazon Bedrock with adaptive thinking (including xhigh), inference profiles, and prompt caching. New ANTHROPIC_AUTH_TOKEN environment variable provides bearer authentication for Anthropic-compatible gateways. pi.dev model catalogs now use If-None-Match revalidation, returning 304 on unchanged catalogs instead of a full download. The outputPad setting is exposed to custom message renderers. Several bug fixes for compaction, branch summaries, unavailable scoped models, and startup context file discovery.

featuremodel-supportnew-model
1mo
06:12
aidev pi

Pi v0.82.0: constrained tool sampling, OpenRouter and Kimi Code OAuth sign-in, session-aware bash

Pi adds constrained sampling for tools with strict JSON Schema, OpenAI Lark, and regex grammar support across OpenAI, Anthropic, Bedrock, Gemini, and Mistral providers. OpenRouter and Kimi Code subscriptions can now be authorized via /login without manual API key configuration. Bash tools receive session metadata (PI_SESSION_ID, PI_PROVIDER, PI_MODEL, etc.), and RPC bash commands stream real-time execution events.

featurecapabilitymodel-support
1mo
16:45
1mo
13:34
aidev pi

Pi v0.81.0: llama.cpp model management, full provider extensions, Qwen Token Plan providers

Pi adds native local model management via llama.cpp (HF search, download, load/unload with live progress), lets extensions register complete pi-ai providers with auth and streaming, ships built-in Qwen Token Plan providers for international and China regions, and expands usage accounting to cover tools, compaction, and branch summaries. A large batch of fixes improves session startup latency, replay reliability, terminal cursor handling, and more.

featurecapability
1mo
22:05
pi Pi v0.80.10: Kimi Coding adaptive thinking fixes v0.80.10 · SIG 2/5
1mo
17:23
1mo
14:40
aidev pi BREAKING

Pi v0.80.8: ModelRuntime, xAI OAuth, Grok 4.5, SDK breaking changes

Major release introducing ModelRuntime — a unified async model configuration and auth facade — replacing the synchronous ModelRegistry and AuthStorage APIs in the SDK. Adds xAI device-code OAuth login, Grok 4.5 with configurable thinking levels, live model catalog refresh via /model and pi update --models, and extension provider dynamic model discovery. Existing SDK consumers must migrate from AuthStorage and ModelRegistry methods to ModelRuntime.

featurebreakingapi-changemodel-support
1mo
16:41
ALL 31 EVENTS →