Context
After a full audit of chat bridges and agent switching across runtimes, OpenCode + kimaki is currently the most capable path for multi-agent Intelligence setups.
Why OpenCode wins for multi-agent
OpenCode has first-class agent switching:
- Named agents in
opencode.json with separate system prompts (build, plan, custom agents)
- Each agent loads different Data Machine context files (SOUL.md, MEMORY.md, etc.)
- Kimaki exposes this via
/agent command in Discord — switch agents at runtime within a channel
kimaki send --agent <name> for programmatic agent selection
Claude Code does not:
- One CLAUDE.md per directory — no agent switching without switching directories or rewriting the file
- cc-connect maps one
[[projects]] entry to one agent config
- No
/agent equivalent — you'd need separate bots or separate projects for different agent personas
What this means
- The
dm-agent-sync plugin already syncs Data Machine context into OpenCode's agent definitions
- When a user creates new agents in WordPress (via
wp datamachine agent create), those could be auto-synced into opencode.json's agent block
- Kimaki's project channels + thread-per-session + agent switching = the most natural Discord experience for Intelligence
Recommendations
- Document this — setup SKILL.md and README should note that OpenCode + kimaki is the recommended path for multi-agent use cases
- Invest in the sync —
dm-agent-sync should auto-populate OpenCode agents from Data Machine agent registry, so creating an agent in WordPress makes it immediately available via /agent in Discord
- Claude Code gap — file upstream or find a workaround. cc-connect could potentially manage multiple Claude Code processes with different
--system-prompt flags, but that's a hack compared to OpenCode's native support
Current state
Both bridges work on the same WordPress backend (same memory, same asks, same state). The difference is purely in agent switching UX:
| Capability |
OpenCode + Kimaki |
Claude Code + cc-connect |
| Agent switching |
/agent command, per-thread |
Not supported |
| Named agents |
opencode.json agent block |
N/A (one CLAUDE.md) |
| Discord integration |
Deep (project channels, threads, worktrees, voice) |
Basic (DM/channel messages) |
| Platform options |
Discord only |
9 platforms |
| Tool depth |
Good |
Better (code editing, permissions) |
🤖 Generated with Claude Code
Context
After a full audit of chat bridges and agent switching across runtimes, OpenCode + kimaki is currently the most capable path for multi-agent Intelligence setups.
Why OpenCode wins for multi-agent
OpenCode has first-class agent switching:
opencode.jsonwith separate system prompts (build,plan, custom agents)/agentcommand in Discord — switch agents at runtime within a channelkimaki send --agent <name>for programmatic agent selectionClaude Code does not:
[[projects]]entry to one agent config/agentequivalent — you'd need separate bots or separate projects for different agent personasWhat this means
dm-agent-syncplugin already syncs Data Machine context into OpenCode's agent definitionswp datamachine agent create), those could be auto-synced intoopencode.json'sagentblockRecommendations
dm-agent-syncshould auto-populate OpenCode agents from Data Machine agent registry, so creating an agent in WordPress makes it immediately available via/agentin Discord--system-promptflags, but that's a hack compared to OpenCode's native supportCurrent state
Both bridges work on the same WordPress backend (same memory, same asks, same state). The difference is purely in agent switching UX:
/agentcommand, per-threadopencode.jsonagent block🤖 Generated with Claude Code