// open-source · self-hosted · Apache 2.0
A directory defines an agent.
One foundationruns the fleetnever leaves homestays auditableruns the fleet.
YokeOS is a Java-native Agent foundation (Agent Harness OS) on your own infrastructure — private, fully auditable, no cloud lock-in. Not another chatbot: the layer that lets a fleet of agents run reliably and be managed. The Phase-1 runtime kernel is being delivered unit by unit.
Three entrances.One engine, one storage.
Drop a directory.Get an agent.
One Directory = One Agent
AGENT.md = frontmatter (runtime config) + a body of instructions. Create via REST, draft from one sentence, or drop the directory into the workspace — live with no restart.
Learn more →LLM Routing
A Provider abstraction unifies mainstream models — agents are vendor-agnostic. Multiple providers coexist via an explicit name → ChatModel map, never type scanning.
Learn more →Self-implemented ReAct Loop
The reasoning engine is implemented in-house — no external agent framework. The LLM picks the tool, the foundation executes and feeds results back. Loop behavior stays fully controllable.
Learn more →Two-layer Memory
Session memory survives restarts; long-term memory is one MEMORY.md file with core/archival partitions, auto-injected into every prompt. Pluggable backends.
Learn more →Tools & Sandbox
Nine built-in tools cover the shortest path; every call passes path/command/domain whitelists and lands in the audit trail. Three extension tiers, zero code first.
Learn more →Notify, Schedule & Web Service
Agents push results via webhook when done; cron runs them on their own — the third trigger source. 18 REST endpoints expose everything, plus a Phase-1 web console.
Learn more →Six capabilities, real enterprise scenarios.
Omnichannel Support
Understands queries, loops over the knowledge base, recalls customer history, connects to CRM — integrated over HTTP.
DevOps Agent
Triages alerts, queries logs, restarts services, recalls past incidents, connects to Prometheus/SSH — triggered via webhook.
Engineering Assistant
Reads requirements, reads and edits code, remembers project conventions, connects to GitHub/CI via IDE plugins.
Knowledge Management
Answers questions, retrieves documents, remembers team conventions, connects to Confluence, embeds in the intranet portal.
Sales Assistant
Assembles customer profiles, calls CRM and business-info tools, recalls preferences — invoked from the sales app.
Data Analysis
Generates SQL, runs queries, draws charts, remembers schema conventions, connects to BI tools and dashboards.
Slow is fast — phase by phase.
Single-node Runtime Kernel
- Six core capabilities
- One directory = one agent, dynamic management, multi-agent coexistence
- Audit tables and whitelist sandbox from day one
- Delivered unit by unit against the reference, equivalent acceptance
Completion & Distributed Foundation
- Knowledge base & semantic memory: ingest, chunk, vector retrieval
- Stateless nodes, externalized state, multi-replica HA
- Platform baseline upgrade (Spring Boot 4 + Spring AI 2.0)
Cross-node Agent Collaboration
- Agent communication substrate with A2A
- Cross-node discovery, delegation, reliable async coordination
Run the fleet on infrastructure you fully control.
From positioning to technical design, the whole initiation document chain is public — every capability has a traceable spec. The Phase-1 runtime kernel is being delivered unit by unit; read along.