Pi coding agent vs OpenCode: same model, headless
All 40 coding runs passed; on DeepSeek, OpenCode used 2.3-3.9× Pi's mean input tokens. With a bad API key, Pi --mode json exited 0 and OpenCode exited 1.
orbi.build · author
Larry 🍀 Orbi
Creator of Orbi, the lights-out software factory for the agent era that turns labelled GitHub Issues into independently reviewed, merged PRs and tagged releases. Every post on this blog is written by him.
OpenCode vs Claude Code: same model, headless
On three small tasks with DeepSeek, both passed every run. Default Claude Code used 1.9-3.0x OpenCode's input tokens; Claude Code --bare used 0.21-0.36x.
Pi coding agent vs Claude Code: same model, measured
With permissions open, both passed every run on two small tasks. Default Claude Code used about 7 to 8 times Pi's input tokens; --bare came close to Pi.
Pi extensions in production: isolate, then write one
How Orbi runs Pi coding agent extensions unattended: a declared --no-extensions allowlist, the D-Bus hang behind it, and two extensions we wrote ourselves.
Pi coding agent worktree: a failed recovery
A surviving branch stalled Orbi's queue after a failed run. How Pi sessions, worktrees and a stable branch help recover work before a pull request exists.
Aider vs OpenCode: context, commits and checks
Compare Aider vs OpenCode 2 by file context, Git commits, tests and permissions. Use the same starting commit to inspect what each agent actually changes.
OpenCode alternatives: who merges and releases?
OpenCode takes an Issue to an open PR and a review, then stops. The OpenCode alternatives below differ in who owns the merge and release after the PR.
Claude Code pricing: Pro, Max and API billing
Compare Claude Code pricing for Pro, Max and API access. Check shared limits, extra usage, API credit exclusions and which account pays before you upgrade.
Cursor vs Claude Code: workflow, plans and usage
Choose Cursor vs Claude Code by workflow, execution environment and billing. Compare current plans, Claude access and a simple task evaluation checklist.
Can an AI agent contribute to open source?
Orbi fixed issues in brig, LLMVault and pyinfra, and maintainers merged all three PRs. On pyinfra our automated review missed a bug and a human caught it.
Claude Code subagents setup and when to use them
Claude Code subagents handle delegated tasks in separate contexts. Configure a focused reviewer, give it clear inputs, and decide when delegation helps.
How to check and reset Codex usage limits
Codex usage limits use five-hour and weekly windows set by plan. Check them on the Usage page or with /status, see when they reset, and keep runs going.
Codex YOLO mode: what runs without approval
Codex YOLO mode skips approvals and the local sandbox. Linux tests show what it allows; the guide also explains Codex auto mode and the full-auto removal.
YOLO mode Claude Code vs auto mode: what I measured
YOLO mode in Claude Code is --dangerously-skip-permissions. I tested it against auto mode and the safer modes on 2.1.287: what runs, what stops injection.
New users waited 1–2 min; long polling made it 10 s
One new user left our preparing page after their workspace was already ready. Long polling and a handful of small fixes cut the wait to about 10 seconds.
Claude Code sandbox: what each setup blocks
Tested on Linux with --dangerously-skip-permissions: the default Claude Code sandbox let four escapes through; strict settings plus a deny rule held.
Ralph loop's code passed 28 tests, then failed review
A Ralph loop reruns one prompt. Mine wrote code that passed all 28 tests, including 13 it never saw, yet a separate review session found a spec violation.
Write a request: Orbi Cloud drafts your GitHub Issue
Write a request in Orbi Cloud: describe a change, review a GitHub Issue draft, answer open questions, and hand it off. See merge controls and trial costs.
Pi agent harness: CLI, AI SDK, Pi Durable vs kill -9
I ran Pi three ways (pi --print, the AI SDK's HarnessAgent, Cloudflare's PiHarness) and killed each mid-task. What came back, and why.
Pi + DeepSeek Flash, 169 merged PRs: 8–16¢ each
In model tokens at DeepSeek off-peak prices, 169 merged deliveries cost a median of $0.082 each. One issue through four Pi sessions; what Pi leaves out.
Claude Code headless mode exits 0 despite denials
In Claude Code headless mode, a claude -p run with denied tool calls still exits 0. Use --output-format json and check permission_denials to catch it.
Searching for Orbi's harness, part 2: guard cost
On eight unseen bugs whose Issue shows only a narrow example, Orbi's original harness failed a third of its runs. Five designs to cut the cost of the fix.
Searching for Orbi's harness, part 1: review gaps
Orbi's review passed two fixes that each broke something. We rebuilt the harness one rule at a time and reran the loop on failed and unseen issues.
GitSpawn and the agent that never asks
A git flaw hit seven coding agents by skipping their approval prompts. Orbi has none to skip, so we shipped a fix, checked who runs git, and reverted it.
DeepSeek coding agent cost per merged PR
See how Orbi uses DeepSeek as an unattended coding agent, what each merged PR costs, why cache reads matter, and how to connect your own API key.
Why Orbi is open source again
Orbi began on Apache-2.0 and switched to the Sustainable Use License on Sep 4. From v0.5.44 it is AGPL-3.0, or the SUL if AGPL is ruled out. Here is why.
The k8e etcd ticket that closed too early
Orbi wrote an etcd test suite for k8e, a 497-star Kubernetes distro. It was rejected, fixed and merged, and the four-phase issue closed after phase one.
How one slash command got its shape
A close look at the name, syntax, regex anchor, and registry behind one in-ticket command, compared with Prow and bors-ng and checked against their source.
The engine waited for an answer users could not give
Our engine paused for a human decision, then asked tenants to use a command they did not have. Three versions of the same mistake in one afternoon.
Seven steps of an Orbi Cloud setup, on video
An 83-second video follows the Orbi Cloud setup from GitHub sign-in to a tagged release and subscription. The final step starts paid Cloud delivery here.
What an autonomous coding agent does when it says no
Most autonomous coding agents stop at a pull request. Here, a second model rejected the diff twice, while the commits and release remain inspectable.
Claude Code in Actions: who presses merge?
Claude Code Action can implement and review pull requests in your runner. It stops before merge and release. Here is the evidence-backed layer for both.
The Docker image made it out on try three
The orbi image reached GHCR and Docker Hub on its third run. The first two failures were build context and Dockerfile paths; the last an invalid tag.