People and agents, building together

ALV is a human-agentic foundry turning product ideas into dependable software

Apache 2.0. We’re onboarding founders and engineers first, by hand.

alv v0.14.13 · repo: aliveness/core · branch: main ● LIVE STREAM · [NORMAL]
PLAN 2 init · 4 chg ● BUILD 2 auto LEARN 1 eval FLEET 3 agents
INITIATIVES & CHANGES
▾ secure-unattended-agent-exec
├ sandbox-runtime ✓
├ unattended-supervisor ✓
└ siwp-git-airgap ✓
▾ alv-tui-dashboard
├ build-agent-fleet ●
└ learn-stage ○
LIVE SANDBOX · alv auto
[Task #102] Implement TDD seam 3/4
agent-alpha · microVM sandbox #1 · worktree isolated
✓ 42 tests passed [folded]
✗ Given a paused worker, When resumed, Then logs replay
→ editing internal/supervisor/resume.go
✓ 43 tests passed
▌
GATES
Ready (DoR) ✓ pass
Done (DoD) ◌ running
FLEET
agent-alpha running
agent-beta running
agent-gamma idle
[←↓↑→] Nav [Tab] Stage [s] Start auto [g] Gate check [?] Help

One loop, from idea to production.

No hundred-page process manual. Three stages in the chat or terminal you already use.

  1. Plan

    Agree on the problem worth solving, how you’ll know it worked, and exactly how the product should behave.

  2. Build

    Agents build it in small, tested steps, working beside you or safely on their own overnight.

  3. Learn

    Ship it, see whether it worked, and keep what you learned for next time.

Then it loops. What Learn records becomes context for the next Plan, so every change starts smarter than the last.

Strict checks keep the agents honest

Specs are the contract

Agents build against behavior your PM can actually read. Review checks every change on two axes: the spec, and your engineering standards.

spec.md
Scenario: Resume a paused worker
Given a worker paused for intervention
When the engineer resumes it
Then the log stream replays from the checkpoint

Gates that don’t guess

Definition of Ready and Done are checked by a compiled gate engine, deterministic and fast, instead of asking a model whether the work looks finished.

alv_gate_check(gate="dod")
tests 43 passed, 0 failed
spec scenarios all covered
evaluated on host, outside the sandbox
result PASS

Agents in a sealed room, on a budget

Unattended agents run in hardware-isolated microVMs with no host credentials and an egress allowlist. Each task is capped on turns, tokens and time, and rolls back to its checkpoint when it fails.

sandbox microVM · no host credentials
egress github.com, registry.npmjs.org, LLM APIs
budget 15 turns · 250k tokens · 45 min
checkpoint baseline saved
✗ 3 consecutive red tests → rolled back, task blocked

Bring your own agent. ALV speaks MCP, so coding agents call typed tools instead of scraping shell output. Use models from Anthropic, Google or OpenAI.

One cockpit, coming soon. Plan, Build and Learn in one keyboard-driven terminal dashboard. Agents keep working when you close it.

Start where you are.

Engineers

Spend your attention on the hard parts. Hand agents sliced, spec’d tasks that run red-green-refactor, then review against the spec, not a vibe.

Product managers

Run /capture-intent and answer one question at a time. You get a proposal now, and behavior scenarios you can actually review later.

Founders

A shorter path from idea to a production-grade product, with engineering discipline built in from the first commit instead of bolted on later.

Built with itself.

ALV is developed with ALV. Every change, from the sandbox runtime to the dashboard above, starts as a proposal, earns its gates and lands with its decision record. Licensed Apache 2.0.

$ ls alv/changes
2026-10-25-alv-tui-dashboard-plan-stage
2026-10-25-alv-tui-dashboard-build-agent-fleet
2026-10-21-security-and-privacy-audit-for-open-sourcing
2026-10-17-monitor-sandboxes
2026-10-12-secure-unattended-agent-execution-siwp-git-airgap
2026-10-12-secure-unattended-agent-execution-sandbox-runtime
2026-10-06-rebrand-av-to-alv-core-cli-engine
…

We’re onboarding the first builders by hand.

We’re starting with founders and engineers. Here’s what happens after you ask:

  1. 1We read and reply to every request.
  2. 2We set up ALV with you on your own repo.
  3. 3We pair on your first change, from proposal to merged PR.