R
Rethink Paradigms
v0.1 · Go · Apache 2.0 · Building in public
Infrastructure for agents that think, persist, and move.
Portable runtimes. Decentralized mesh. Replayable intelligence pipelines. Running on your hardware, under your control.
Get Started
OUR THESIS
The Problem
Containers were designed stateless — kill and recreate, state in the DB. AI agents broke this. An agent is a persistent process that writes files as it works; the filesystem IS the state. Nothing in the existing stack is shaped for this. And the current approach — wrapping models in API calls and hoping for the best — produces systems that are fragile, unverifiable, and expensive.
The Insight
The winning architecture constrains the model into narrow, validated corridors. Wraps it in deterministic pre/post-processing. Treats context as a managed resource. And runs on infrastructure the user owns — not a cloud provider's shared tenancy. The agent needs a body it controls.
OPEN SOURCE
Mesh
Portable Agent-Body Runtime
Containers were designed stateless — kill and recreate, state in the DB. AI agents broke this. An agent writes files, installs packages, builds context. The filesystem IS the state. Mesh gives any AI agent a persistent body that survives restarts, moves between machines, and runs on any substrate.
STATUS
v0 shipped
STACK
Go · OCI · Substrate Plugins
SHIPPED
snapshot · restore · clone · status · list · inspect · prune
LICENSE
Apache 2.0
COMMERCIAL
Agent Bodies
Governance Interface for Agent Compute
If Mesh is Git, Agent Bodies is GitHub. Type in plain English, AI proposes actions, you confirm. Everything reversible. Preview Render replaces confirmation dialogs — see consequences before they execute. The body card dims, dependencies pulse, a countdown starts with undo.
STATUS
In design · 11 UX sessions
STACK
React · WebSocket · MCP
INTERFACE
Prompt · Canvas · Rail
APPROACH
2026 anxiety-free computing
INTELLIGENCE
Agent Lab
Replayable Intelligence Pipelines
Break workflows into modular nodes with defined I/O schemas. Checkpoint every step. Rerun from any point. Let specialized agents optimize each node. The design is the source code; the code is the compiled output. Chemistry metaphor: stable apparatus, disposable experiments.
STATUS
Architecture v0.1 planned
STACK
Python · Pydantic · MCP · SQLite
NETWORK
TS device routing in-memory
APPROACH
Laboratory, not a task runner
ARCHITECTURE
L4 LAYER
Agents · Substrate Plugins · Workflows
L3 RUNTIME
25MB daemon · Portable agent bodies
L2 STATE
Checkpoint · Fork · Resume · Snapshot
L1 INTERFACE
CLI · Web · MCP · Native
TRACTION LOG
1,583
SESSIONS
51,496
MESSAGES
12
PIPELINES
Ready to rethink how agents run?
Join the waitlist for Agent Bodies, or explore Mesh — our open-source agent runtime.
Join Waitlist