Joined December 2024
- Tweets 432
- Following 38
- Followers 440
- Likes 908
18 Photos and videos
Pinned Tweet
May 23
pi-cursor-sdk uses your Cursor subscription usage allowance. No additional API costs.
Bug-fixes for mid-turn steering and edge cases coming later today.
What do you think of Composer 2.5 so far?
May 22
Released pi-cursor-sdk v0.1.16 🚀
Cursor still uses its own harness tools.
But pi-cursor-sdk v0.1.16 adds the missing bridge: Cursor agents can now reach active pi extension tools and other pi-side support through local MCP, while pi keeps the tool cards, confirmations, history, aborts, and diagnostics.
Try Composer 2.5 in pi today. What will you build?
pi install npm:pi-cursor-sdk
github.com/fitchmultz/pi-cur…
8
4
50
12,838
Jun 13
The annual Max coding plan I bought for a few hundred dollars is one of the better decisions I've made all year.
Intelligence should be open, accessible, and ready to build with, empowering every developer, everywhere.
GLM-5.2 is now available to all GLM Coding Plan users, including Lite, Pro, Max, and Team plans.
docs.z.ai/devpack/latest-mod…
As our new flagship model, GLM-5.2 delivers powerful coding capabilities, usable 1M-context support, and continued strengths in long-horizon tasks.
API and Chatbot services will launch next week. The model will also be officially open-sourced next week under the MIT License.
The future of AI is open, and it belongs to the people.
10
1
145
21,914
Jun 11
Cursor's agent loop 🤝 pi's minimalism extensibility
I've generated 2.2B tokens with pi-cursor-sdk so far. Composer 2.5 is hyped a lot on X but it still feels slept on.
Composer 3 when?
Jun 10
pi-cursor-sdk is now at v0.1.42. Last time I posted it was v0.1.31.
It lets you use Cursor’s agent loop inside pi while keeping pi’s terminal UI, sessions, tools, and workflows.
Since then:
- Cursor tool activity now shows as readable pi cards
- Cursor-spawned subagents are visible in the UI
- shell output streams while Cursor works
- releases are now tested on macOS, Linux, and Windows
Install:
pi install npm:pi-cursor-sdk
github.com/fitchmultz/pi-cur…
358
Jun 10
pi-cursor-sdk is now at v0.1.42. Last time I posted it was v0.1.31.
It lets you use Cursor’s agent loop inside pi while keeping pi’s terminal UI, sessions, tools, and workflows.
Since then:
- Cursor tool activity now shows as readable pi cards
- Cursor-spawned subagents are visible in the UI
- shell output streams while Cursor works
- releases are now tested on macOS, Linux, and Windows
Install:
pi install npm:pi-cursor-sdk
github.com/fitchmultz/pi-cur…
3
1
57
6,341
Jun 11
There's an SDK issue with Fable BTW. I cannot get it to work and it seems out of my control. I'll keep an eye on it and post an update when something changes.
146
Jun 3
Shipped pi-agent-browser-native v0.2.41 🚀 ( v0.2.42)
Exa joins Brave — dual-provider web search for pi agents (`auto` prefers Exa when both keys exist).
Set preferredProvider: "brave" in config to flip default.
429 recovery config/Chrome doctor hardening.
pi install npm:pi-agent-browser-native
Brave or Exa default?
github.com/fitchmultz/pi-age…
178
Jun 3
I highly recommend attending some @cursor_ai workshops. Even if you aren't a current Cursor user I'd wager you'd learn something useful.
Jun 3
Join new workshops here and see past ones
cursor.com/workshops
1
2
508
Jun 3
Shipped pi-agent-browser-native v0.2.40 🚀
Agents can now search the live web through Brave as a native pi tool. I've seen a major uptick in productivity by adding search.
Why I like this:
- no search-engine page scraping just to find URLs
- no generic web_search naming collision
- no API key in chat/logs/tool output
- project config only allows safe env refs
- release-smoked on macOS, Ubuntu Linux, and native Windows
Which search engine should I support next?
github.com/fitchmultz/pi-age…
8
6
78
4,589
Jun 3
Install/update:
pi install npm:pi-agent-browser-native
Get a Brave Search API key:
api-dashboard.search.brave.c…
Brave currently gives free monthly Search API credits, which is plenty for light agent use.
Setup docs:
github.com/fitchmultz/pi-age…
297
Jun 2
I adopted Crabbox to test and validate my pi extension projects on different platforms.
The following extensions should work on Windows (native PS), Linux (inc. WSL2), and MacOS:
- pi-cursor-sdk v0.1.31
- pi-codex-goal v0.1.21
- pi-agent-browser-native 0.2.39
- pi-oracle 0.7.5
Please open a GitHub issue if you have any issues.
github.com/fitchmultz
May 13
Crabbox 0.13.0 is live
🧪 Modal sandbox runs
🧼 Full resync for stale workdirs
🪟 Native Windows script preflight support
🔧 Clearer SSH/sync failure hints
Been using that for almost every PR now.
github.com/openclaw/crabbox/…
2
1
33
4,243
Mitch Fultz retweeted
Jun 2
OpenAI-compatible endpoints get you access to Cursor models.
pi-cursor-sdk keeps Cursor’s agent loop as the runtime.
That’s the distinction: access vs integration.
1
3
257
Jun 2
OpenAI-compatible Cursor endpoints are great if you bounce between generic clients.
But if you primarily use pi, pi-cursor-sdk is the better abstraction: keep Cursor’s top-tier agent loop intact, then add pi-native extentions, tools, sessions, context, and replay UI via an MCP bridge.
github.com/fitchmultz/pi-cur…
2
18
1,585
Jun 2
OpenAI-compatible endpoints get you access to Cursor models.
pi-cursor-sdk keeps Cursor’s agent loop as the runtime.
That’s the distinction: access vs integration.
1
3
257
Jun 2
GPT-5.5 is a powerhouse for nearly everything, except writing. I can immediately tell when something is written by GPT-5.5. Grok can write human-sounding text so much better.
Anyone else feel the same? What do you use for writing normal sounding docs and such?
2
4
862
Jun 1
I've been adding Windows (PS) and Linux testing to each of my pi extensions the past 2 days. Soon I'll have a repeatable framework done and be able to provide better cross-platform support for pi-cursor-sdk, pi-codex-goal, pi-oracle, pi-agent-browser-native, and more. Updates soon to come.
6
600
Mitch Fultz retweeted
May 31
Your coding agent spawns a sub-agent.
Sub-agent runs 8 minutes. Main agent wakes up —
cache is cold. You just paid full input token price. 🤑
The issue: OpenAI's in-memory cache TTL is 5–10 min.
Sometimes 1 hour if GPU memory isn't busy.
Unreliable for any sub-agent workflow.
The fix: `prompt_cache_retention: "24h"`
Offloads KV cache.
24-hour deterministic TTL. No extra cost.
For pi, set env var to `PI_CACHE_RETENTION=long`
x.com/i/status/2060843228618…
May 30
Analyzed 16K GPT-5.x API calls across 4 months.
Cache hit rate: 92-95%.
Your subscription isn't draining because of cache misses. The cache works exceptionally well. Misses are rare and explainable. Here's what's actually happening. 🧵
6
4
33
2,752
Mitch Fultz retweeted
May 31
Use the included /create-goal prompt template instead of writing a goal by hand. Agents write better goal completion contracts than humans do because they can expand a plain task into outcome, verification, constraints, iteration, audit, and blocked-stop requirements before calling the create_goal tool.
`/create-goal insert task and requirements here`
The template follows the Codex goal-writing practices from:
developers.openai.com/codex/…
developers.openai.com/cookbo…
2
13
792
May 31
Agents write better goals than humans.
Shipped pi-codex-goal v0.1.19 with a new best-practice slash command:
/create-goal fix this bug, test it thoroughly, no tech debt
It turns a messy task into a strict Codex-style completion contract, then creates the goal.
Also fixed long-session reliability:
- context overflow/session bloat
- compaction edge cases
- transient upstream retry errors
pi install npm:pi-codex-goal
github.com/fitchmultz/pi-cod…
1
5
124
4,790