Turbo Flow version history
Turbo Flow shipped nine tagged releases between September 2025 and October 2026. The short version: a year of packaging an ever-bigger agentic development environment, a March 2026 peak of 215+ MCP tools and 60+ agents, then a deliberate retreat — September 2026's v5 removed the orchestration stack and kept the governance. The full reasoning is in why Turbo Flow became Turbo Rig.
Published Oct 7, 2026 · Updated Oct 8, 2026
The timeline
| release | tag | date | what it was |
|---|---|---|---|
| v1.0.1 | v1.0.1 | 2025-09-21 | first public alpha — "Turbo-Flow Claude": a packaged Claude Code environment for Codespaces, DevPods and cloud shells |
| v1.0.2 | v1.0.2 | 2025-10-17 | alpha iteration on the install chain and workspace bring-up |
| v1.0.4 | v1.0.4 | 2025-12-24 | alpha — environment hardening before the 3.x line |
| v1.0.5 | v1.0.5 | 2026-01-05 | last of the 1.x alphas |
| v3.0 | v3.0 | 2026-03-07 | the Claude Flow era at scale — 175+ MCP tools, multi-agent workflows |
| v4.0 | v4.0 | 2026-03-09 | the Ruflo migration: Claude Flow → Ruflo v3.5, 215+ MCP tools, 60+ agents, Beads memory, GitNexus, worktrees |
| v5.0-preview | v5.0-preview | 2026-09-22 | the rig era begins: rig-lite lands (gate.sh, constitution, memory, self-tests); Turbo Rig private beta opens |
| v5.1.0 | v5.1.0 | 2026-09-28 | rig-lite becomes the complete portable rig: nine files, worktrees, PR-verdict mode, sweep, automations pack |
| v5.2.0 | v5.2.0 | 2026-10-07 | the lean pivot: 68 → 35 files, AGENTS.md as the single source of truth, the v4 product half removed (recoverable in tags) |
The four eras
Turbo Flow started as packaging: take Claude Code, add the pieces it didn't ship with — context loading, memory, multi-surface setup — and make it come up identically on a Codespace, a DevPod, or a cloud shell. The alphas were install-chain work: fewer steps, fewer ghost packages, a workspace that built itself.
v4.0 (March 2026) was a complete rebuild around Ruflo v3.5 — Claude Flow after 5,800 commits and a rename. The numbers from its release notes: setup steps 15 → 10, MCP tools 175+ → 215+, 60+ agents, Beads for cross-session memory (git-native JSONL), GitNexus for codebase-aware agents, native git worktrees for per-agent isolation, and 3-tier model routing that cut model spend by roughly 75%. This is the version the search index still remembers — and the version v5 would mostly delete.
September 22, 2026: v5.0-preview introduced rig-lite — a fail-closed cross-model review gate
(gate.sh), a written constitution, git-versioned memory, and a self-test suite — while the
v4 environment stayed below the fold. v5.1.0 graduated the kit from four files to a complete portable
rig: wt.sh worktrees, PR-verdict mode (GitHub as the state machine), --sweep
for every ungated open PR, and a nine-task automations pack. 166 self-test checks, all passing.
October 7, 2026: Turbo Flow became what it always was under the hood — the rules layer. One product
(rig-lite), one source of truth (AGENTS.md), 68 → 35 files, a 1.8 MB working
tree. The v4 install chain, demo and era docs came out in three gated strips (PRs #62–#64); the whole
v4 era stays recoverable in the v1.0.1 → v4.0 tags. 252 self-test checks, all passing —
grown by the release itself.
Why keep the old tags at all?
Because the history is the product's evidence. The v4 tags document a real system that really ran — and
its deletion is the argument for v5. Anyone can check out v4.0 and count the 215 tools, then
check out v5.2.0 and find 35 files that govern any harness. That contrast, not a marketing
claim, is the pitch. The same reasoning drives
orchestration vs governance and the
research notes.
Where it goes next
Turbo Flow stays the open-source rules layer. The integrated successor — gate, constitution, memory and automations as one always-on system — is Turbo Rig, currently in private beta.