DevConsensus
Terminal agent · Open source · HN opinions Nov 2025 – Nov 2026

Pi: what developers actually say

We read every Hacker News comment from Nov 2025 – Nov 2026 that gave a real opinion on Pi. 338 of the 423 opinions come from people who say they use it themselves.

84%approval, all opinions
80%approval, last 90 days
#1of 22 tools ranked
335different developers

Developers give Pi 84% approval across 423 opinions from 335 authors, and 338 of those opinions come from firsthand use. Customization drives most of the support: 150 positive mentions against 4 negative, mostly for extensions, multi-provider support, and subagent tooling. Approval over the last 90 days is 80%, slightly below the overall figure. Monthly approval fell from 96% in April 2026 to 74% in September, then recovered to 83% in October while comment volume stayed high.

Pick it if

  • Want to build your own extensions and subagent tooling
  • Need multi-provider or local open-weight model support
  • Value an open-source, vendor-agnostic terminal harness

Look elsewhere if

  • Need a lean install with few dependencies
  • Want minimal setup and a short learning curve
  • Are wary of telemetry and movement toward hosted services
Visit Pi ↗

Strengths and weaknesses

Each row counts opinions that mentioned the topic. Bars to the right of the line are praise, bars to the left are complaints.

PraiseComplaints
Customization
4
150
UX
7
36
Cost & pricing
5
29
Reliability
8
25
Autonomy
6
20
Open source
17
Speed
3
14
Privacy & local use
1
13

How opinion has moved

050100Jan '26Mar '26May '26Jul '26Sep '26Nov '26
Monthly approval (line) and number of opinions (bars). Months with fewer than 3 opinions are left out.

What developers say

Praise

Chose Pi over Claude Code for being open source, vendor-agnostic, and extensible; building subagent tooling in it.HN comment · Nov 1, 2026
Switched from Claude Code to Pi two months ago, praising its multi-provider support, visibility, and minimal bugs.HN comment · Nov 1, 2026
Says Qwen coding agents also work well in Pi, using only local open-weight models.HN comment · Nov 1, 2026
Has settled on Pi as part of their setup alongside Zed.HN comment · Nov 1, 2026
Uses bare Pi with self-made extensions; likes that it waits for answers as the next prompt.HN comment · Nov 1, 2026
Used the Pi harness to drive a game decompile and WASM port that impressed the author.HN comment · Oct 31, 2026
Says they personally like Pi among open-source coding harnesses and use it alongside Claude Code.HN comment · Oct 27, 2026
Has tried several coding harnesses and strongly prefers Pi.HN comment · Oct 27, 2026

Complaints

Says Pi ended up a bulky mess, and its Homebrew install stalled after a very long dependency build.HN comment · Oct 30, 2026
Uses Pi, but has no faith in its security or maturity after dealing with its CEO.HN comment · Oct 25, 2026
Is trying to add MDM support to the Pi harness, and says it is not going well.HN comment · Oct 18, 2026
Calls Pi tinkering for its own sake; tried it for a few weeks and found it not worth the effort.HN comment · Oct 2, 2026
Found Pi a bit tough to use at first.HN comment · Oct 1, 2026
Dislikes where Pi went, saying it sold out and dropped its principles; used it only in the past.HN comment · Sep 30, 2026
Tried Pi and did not like it.HN comment · Sep 29, 2026
Says Pi's agent pushed an unwanted update to a published crate during a trial, prompting them to add sandboxing.HN comment · Sep 29, 2026

Summaries are paraphrased. Each links to the original comment.

Pi compared

Best Pi alternatives, according to developers who switched →

FAQ

Is Pi worth it?

Most developers say yes. It holds 84% approval across 423 opinions, and 54 users said they preferred it to Claude Code. Complaints focus on dependency bloat, install trouble, and security maturity.

Is Pi better than Claude Code?

54 people preferred Pi over Claude Code, and 6 moved the other way. OpenCode is the main competitor: 35 preferred Pi to it, while 11 chose OpenCode instead.

Does Pi work with local models?

Yes. Privacy and local use drew 13 positive mentions against 1 negative. One user reported running Qwen coding agents entirely on local open-weight models.