Private beta · Apache-2.0 OSS planned

Run N parallel autonomous AI engineers.

Coordinator-led sprints. Cross-family review. The harness we built to ship our GenAI platform — now opening to your team.

How SessionFleet coordinates parallel autonomous AI engineersA coordinator session fans out to parallel worker sessions in isolated worktrees — Claude Code and Codex — each passing a plan-review gate and a commit-review gate. Plans and commits are reviewed by the other AI family, then merged in dependency order.Coordinatorcross-family reviewWorkerClaude CodeWorkerCodexWorkerClaude Codeplan reviewcommit reviewMerge in dependency order

Single-agent AI dev plateaus

Today’s AI coding tools are single-agent: one chat, one editor, one PR at a time — and the reviewer is the same model that wrote the code.

One session at a time

A single chat, a single editor, a single PR in flight. Throughput is capped at one agent’s pace, no matter how fast that agent is.

The reviewer is the writer

When the model that wrote the code also reviews it, self-bias goes unchecked. You get plausible diffs, not independently-validated ones.

No PM / Eng / Design loop

Specs live in someone’s head. With no structured hand-off from intent to implementation, scope drifts and rework piles up.

There’s a ceiling. We hit it. We built past it.

Coordinator + N workers + cross-family review

SessionFleet runs many autonomous sessions in parallel, coordinated like a sprint, with every plan and every commit reviewed by a different AI family.

Parallel autonomous sessions

Run N isolated worktree sessions at once — Claude Code workers, Codex workers, or both — instead of one chat at a time.

Cross-family review

Claude writes, Codex reviews. Codex writes, Claude reviews. Independent perspective from a different model family closes the self-bias gap, at plan and at commit.

PM / Engineering / Design spec authoring

Every functional spec runs cross-functional deliberation — product, engineering, and design perspectives — before a line is written.

Kanban / sprint coordination

A coordinator session decides what to pull next, sequences parallel-safe work, and merges in dependency order — visible on a board.

The discipline underneath

Speed without discipline is just faster mistakes. SessionFleet bakes in the review and risk primitives staff engineers actually trust.

Tier 1 / 2 / 3 risk classification

Every change is classified by risk. Docs are cheap; auth, migrations, and security paths get the deepest scrutiny — automatically, by path and content.

R3 hard cap + disagreement log

Review rounds are bounded at three. At the cap, work ships clean or ships with an auditable disagreement log — never an infinite loop, never a silent override.

PM + Engineering + Design spec authoring

Specs are triangulated across three perspectives before implementation, so intent survives the trip to production.

Library-currentness gate

Before any framework-specific change, the harness verifies behavior against current docs — not the model’s training memory, which lags.

We build on it every day

We build sessionfleet’s multi-tenant RAG 2.0 and agentic GenAI platform with SessionFleet — every line spec’d, written, reviewed, and merged under it. The harness isn’t a side project; it’s how the platform exists. Real throughput and cost numbers are coming as we stabilize our internal measurement.

Roadmap

Where SessionFleet is, and where it’s going. Public and honest.

Now

Private beta

  • Coordinator + parallel sessions + cross-family review, in daily use
  • Join the waitlist for early access

Next · 4–6 weeks

Open-source GA

  • Apache-2.0 release, public repository
  • Claude Code + Codex marketplace install

Q3

Hosted add-on

  • Managed coordinator + dashboards
  • Team visibility into N parallel sessions

Q4

v1.0

  • Provider-agnostic workers
  • Stabilized review + cost primitives

Frequently asked

Is it really open source?

Yes — the planned license is Apache-2.0, and the repository goes public at GA. Today it’s a private beta; join the waitlist for early access.

Does it lock me into Anthropic or OpenAI?

No. SessionFleet coordinates Claude Code and Codex workers today and is designed to become provider-agnostic over time. Cross-family review is the whole point — a layer no single vendor would ship.

What about Cursor, Devin, or Aider?

Those are workers or single agents. SessionFleet sits above them: it coordinates many sessions and adds independent cross-family review. We orchestrate; we don’t replace your editor.

What’s the security model?

Risk-tiered paths, sensitive-path gates, bounded review rounds with audit trails, and scoped approval tokens. The full SECURITY.md ships with the public repository at GA.

How do I contribute?

Contribution guidelines open with the public repository at GA. Join the waitlist and we’ll let you know the moment it’s live.

Join the beta waitlist

SessionFleet is in private beta. Leave your details and we’ll reach out as we open access — no spam, no sharing.

Beta waitlist opens shortly — check back soon.

SessionFleet

The harness for running N parallel autonomous AI engineers. Open-source (Apache-2.0) release planned.

© SessionFleet · Apache-2.0 release planned · private beta

Mode
Tone
Accent