Interactive REPL Runtime for AI Agents
The Interactive REPL Runtime for AI Agents
Ditch static graphs and brittle prompt chains. Run, steer, and introspect autonomous agents inside a persistent live runtime — spawn sub-agent swarms, compile skills in runtime, and attach via REPL in real time.
zea-repl — session_8f92a (live attached)
ONLINE
# 1. Inspect live agent swarm state
> zea.swarm.status()
%{lead: "architect", subagents: ["db-optimizer", "security-auditor"], status: :running}
# 2. Steer execution in flight without losing context
> zea.steer("prioritize concurrent index creation for postgres migrations")
✓ Injected into live context. Agent adapting plan in runtime...
# 3. Dynamic skill compilation at runtime
> agent.skills.compile(:pg_index_advisor, &Database.analyze_indexes/1)
✓ Skill :pg_index_advisor compiled & hot-reloaded into cluster memory
⚡
REPL-Driven Runtime
Persistent execution environment. Agents write code, run tools, and introspect their trajectory inside a live interactive REPL.
Runtime
@zea.cl/soma-sdk
🎯
Live Steering & Introspect
Attach to any running session in real time. Inspect live memory, steer decisions mid-flight without losing context.
Runtime
@zea.cl/soma-sdk
🧠
Dynamic Skills & Memory
No more static prompts. Agents compile, register, and hot-reload new tools and reusable skills in runtime as they solve tasks.
Runtime
@zea.cl/cerebelum-sdk
🛡️
Kernel Sandboxes
Run agent-generated code safely. Isolated Linux environments with kernel permissions, git access, and secure bash execution.
Runtime
@zea.cl/soma-sdk
💬
Real-time Messaging
WebSocket-native event and message streaming between users, agents, and sub-agent swarms with typing indicators and channels.
Runtime
@zea.cl/synapse-sdk
🔐
Unified Identity (OAuth 2.1)
Centralized identity and fine-grained permissions for humans and agents. OAuth 2.1 PKCE, JWTs, and org-scoped API keys.
Runtime
@zea.cl/auth
Orchestration as Code. Not Static Graphs.
Forget rigid canvas tools and static graph pipelines. Build adaptive systems with standard async code, live REPL steering, and runtime skill compilation.
01
Orchestration as Code
Define and spawn sub-agent swarms using pure idiomatic code (async/await, loops, events). No visual drag-and-drop canvas, no YAML hell.
02
Live REPL & Steering in Flight
Connect to any running agent from CLI or code. Inspect active state, watch sub-agents execute in parallel, and steer decisions without restarting.
03
Runtime Skill Compilation
When an agent discovers a reusable pattern or fixes a bug, it compiles and hot-reloads the new skill into memory for immediate use across the cluster.
04
Kernel-Isolated Execution
Every agent operates in a sandboxed Linux environment. Agents can clone repos, install npm/pip packages, and run tests with zero risk to your host.
Start building in the REPL today
One cluster. Live introspection. Self-improving agent swarms.