Skip to content

The team

Four humans,
one agent runner.

A small human team that ships across the catalogue, plus a single agent runner per session. The agent stack is one tool, not eight personas.

Humans

Four people. Each one owns a real surface.

🪶

Leif

@0xLeif

Owner / Lead Architect

Creator and project lead of CorvidLabs. Architecture, product, and team decisions all land here. Currently building corvid-chat + Merlin.

corvid-chat Merlin fledge spec-sync
🎨

Tofu

Designer / Co-founder

Co-founder and creative direction for the CorvidLabs aesthetic. Brand, UI, NFT collections, the crow mark.

Brand UI NFTs
⚙️

Kyn

@Kyntrin

Developer

Kotlin / Ktor backend developer. Keeps the team honest and holds shipped work accountable.

Kyntor kt-algochat KMP
🏗️

Gaspar

@0xGaspar

Infrastructure

Infrastructure focus. Rust, Podman, Kubernetes. Where the self-hosted runners and prod servers come from.

Infra Rust K8s

The agent, honestly

Sessions are launched per task. There's no always-on roster of specialized agents. That's aspirational shape we don't claim. Here's what's actually been driving the work:

  1. Historically Claude Code

    Worked. Productive. But the daily-driver budget was metered, and the agent was something CorvidLabs paid to use, not something it owned.

  2. Today Merlin

    Sessions run on Merlin, the in-house runner: it talks to providers over their APIs directly, with no CLI to shell out to. That includes Claude. Merlin can still call Claude models straight over the Anthropic API. What changed at v0.4.0 is that Claude Code stopped driving CorvidLabs' agent sessions; Merlin has carried that work since. The team still uses Claude Code for its own day-to-day development. This is only about what runs the agent. Per-task, not always-on.

  3. Next Merlin 1.0

    Merlin is at v0.11.0: sub-agents, long-task durability, the audit chain, and spend caps have all landed. What's left is 1.0 polish: install, onboarding, and reliability, then a public release.

Read the runner: Merlin. Read the strategy behind it: the dependency chain.

How we work

Humans plan and steer. The agent runner executes per task. Git worktrees keep parallel work isolated; spec-sync keeps the contract honest; fledge runs the same verify lane on every commit.