We build tools for developers
who code with AI.
Open source. Ship-ready.
From a production multi-agent orchestration system built on Claude.
Open Source
Developer tools extracted from our internal orchestration system.
specfirst
Confidence-scored development plans. Know what you're building before you build it.
guardian
Pre-tool-use policy gate with audit trail. Intercepts and evaluates every tool call.
voicekeeper
AI detection and voice matching. Your content, your voice.
dispatch
Lightweight MCP task server. Coordinate agents without the overhead.
dreamwatch
Overnight autonomous execution with budget caps and git isolation.
xfetch
Social media content extraction. Get the data without the noise.
dual-auth
Firebase Auth + API key dual authentication with encryption utilities.
rate-gate
Sliding window rate limiter with pluggable backends.
poll-hooks
Async work-polling with claim semantics and lifecycle hooks.
Public Interfaces
The open-source tools represent production-hardened primitives extracted from a larger orchestration system. The coordination protocol, identity model, and decision framework remain private.
CacheBash
Async dispatch infrastructure for AI agent networks.
MCP relay for multi-agent coordination. Typed messaging, transactional task claiming, and session persistence. Built on Firestore + Cloud Run. CLI-based orchestration infrastructure.
Learn moreThe engine behind the studio.
Controls
Context runaway mitigation via threshold-based self-handoff
Pre-tool-use policy gating on all tool invocations
TTL-based relay expiration with dead-letter handling
Identity-scoped execution boundaries
Static token ceilings per session
Explicit WORK vs CONTROL task taxonomy
How we build
Ship-ready, not ship-someday.
Every tool works on install. No configuration mazes. No "coming soon" features in the CLI. If it's in the README, it runs.
Subtract to ship.
We cut features until what remains is clean and obvious. Enterprise-grade means powerful AND simple. Not powerful because it's complex.
Bounded autonomy. Not blind autonomy.
Every tool enforces explicit constraints—context limits, token ceilings, identity boundaries, policy gates. Autonomy is allowed only within defined envelopes. We instrument failure modes before we scale execution.
The tools are the public interfaces. The orchestration layer runs beneath.
github.com/rezzedai →