This project contains a comprehensive OpenCode configuration setup designed to enhance AI-assisted development workflows. OpenCode is an open-source AI coding assistant that provides intelligent code completion, refactoring capabilities, and seamless integration with various development tools and platforms.
The configuration includes carefully selected plugins, commands, and agents that extend OpenCode's functionality to support complex development tasks, improve code quality, and streamline development workflows. This setup is particularly focused on providing enterprise-grade features while maintaining flexibility for personal and team use cases.
OpenCode serves as a powerful alternative to traditional IDE-based AI assistants, offering features such as context-aware code generation, multi-file refactoring, intelligent code search, and integration with popular development tools like GitHub, Jira, and Figma. The configuration files in this project customize OpenCode's behavior to match specific workflow requirements and preferences.
This repository is being simplified around three directions:
-
Move skills into Skillless. Skills and skill presets are moving to skillless, so this repository can stay focused on OpenCode configuration instead of becoming a mixed skill/config bundle.
-
Remove Anthropic-specific OpenCode paths. The
anthropicprovider andopencode-with-claudeplugin are being removed from this setup as a personal workflow decision. Claude models currently run better in Claude Code than in OpenCode with Oh-My-OpenAgent, and Claude Code gives smoother control for Claude-centric work. -
Centralize MCPs through MCPProxy and prefer CLIs where possible. Most direct MCP configuration is being removed so multiple coding agents do not each maintain their own MCP setup. Shared MCPs now live behind MCPProxy, using the
retrieve_toolsrouting mode to search tools on demand and lazy-load only the tools matching the current keyword or task instead of injecting every MCP tool into context. Over time, some MCPs are also being replaced with CLI-based presets from Skillless, such as the CLI preset list. The goal is to replace high-token MCP flows such as Exawebsearchand Context7 with CLI workflows where they save tokens and perform better.
The Makefile provides essential commands to manage the OpenCode configuration:
| Command | Description |
|---|---|
make sync |
Sync OpenCode configuration into ~/.config/opencode/. Copies AGENTS.md, JSON config files, and mirrors root ./agents/ and ./commands/ directories if they are present |
make help |
Display available targets and their descriptions |
Workflow:
- Run
make syncto copy configuration files and any root-levelagents/orcommands/files to system locations - Use
make helpto see all available commands
The current repository stores project-scoped commands under .opencode/commands/; there are no root-level agents/ or
commands/ directories at the moment.
This configuration integrates multiple AI model providers to offer a diverse range of capabilities, from lightweight fast responses to deep reasoning tasks. The provider setup is designed to balance cost-effectiveness with performance, utilizing both free and premium models across different use cases.
Default Model: openai/gpt-5.4 (configured in opencode.json)
Small Model: openai/gpt-5.4-mini (configured in opencode.json)
The following AI providers are documented for this setup because they are used by model routing or explicit provider configuration:
GitHub Copilot
GitHub Copilot is used as a routed model provider for selected task categories and fallbacks. The current
oh-my-openagent.json configuration references github-copilot/gemini-3.1-pro-preview,
github-copilot/gemini-3-flash-preview, and github-copilot/gpt-5-mini.
These routes give the setup Gemini-backed high-effort coverage plus a lightweight Git-focused model path.
OpenCode
OpenCode's built-in model hub offers several models optimized for different task types. The configuration keeps
opencode/big-pickle available as a large-context fallback for selected agents and task categories while filtering
paid Zen models from routine provider selection. These models provide a reliable, cost-effective foundation for
high-frequency workflows.
Z.ai Coding Plan
Z.ai provides access to advanced GLM models. In this configuration, the GLM family is used for strategic reasoning,
planning, visual fallback coverage, and multimodal fallback coverage, with zai-coding-plan/glm-5.1,
zai-coding-plan/glm-5v-turbo, and lightweight 4.x variants such as
zai-coding-plan/glm-4.5-air assigned across specialist agents and task categories.
xAI (Grok)
xAI provides the Grok family of models. In this configuration, non-LLM Grok Imagine image and video generation models
are filtered via provider.xai.blacklist so text and coding workflows only see suitable LLM choices.
oMLX
oMLX is configured as a local OpenAI-compatible provider served from http://127.0.0.1:8080/v1. It exposes two
manually named MLX-hosted Qwen variants in provider.omlx.models, giving this setup a local inference path alongside
hosted providers.
OpenAI
OpenAI provides direct API access to GPT-5 family models including openai/gpt-5.5, openai/gpt-5.4,
openai/gpt-5.4-mini, openai/gpt-5.4-nano, and openai/gpt-5.3-codex. These models serve as primary and fallback
models for several agents including hephaestus (primary: openai/gpt-5.4 medium, ultrawork: openai/gpt-5.5
medium), oracle (primary: openai/gpt-5.5 high), momus (primary: openai/gpt-5.5 xhigh), and
multimodal-looker (primary: openai/gpt-5.5 medium). Authentication is handled via the opencode-openai-codex-auth
plugin.
NVIDIA
NVIDIA provides access to models hosted on the NVIDIA AI platform. This configuration uses
nvidia/minimaxai/minimax-m2.7 as the primary model for explore and librarian, with additional fallback use in
atlas, sisyphus-junior, quick, unspecified-low, and writing routes.
DigitalOcean
DigitalOcean's AI platform provides access to open and partner models used heavily by this configuration. It serves
digitalocean/kimi-k2.6 as the primary model for sisyphus, atlas, and sisyphus-junior,
digitalocean/kimi-k2.5 for writing and fallback routes, digitalocean/glm-5 for visual and high-effort fallbacks,
and digitalocean/deepseek-3.2 as the fallback for explore and librarian. Premium anthropic-* and openai-*
DigitalOcean models are filtered via provider.digitalocean.blacklist, while open-source openai-gpt-oss-* variants
and other open or partner model families remain available.
The provider configuration also uses blacklist rules to keep model selection focused and predictable.
OpenCode Zen blacklist (provider.opencode.blacklist)
This blacklist is maintained to filter paid OpenCode Zen models from the general OpenCode provider roster while keeping
free-tier models available. It is synchronized via /blacklist-sync and currently filters 41 paid Zen models across
families including Claude (Fable, Opus, Sonnet, Haiku 4.x), GPT (5.x, Codex, Nano), Gemini (3.1 Pro, 3 Flash), Grok
Build, DeepSeek V4, GLM 5.x, MiniMax M2.x, Kimi K2.x, and Qwen 3.x so routine workflows stay on the free/default
OpenCode path.
xAI blacklist (provider.xai.blacklist)
This blacklist is maintained to keep the xAI model list focused on text and coding workflows. It is synchronized via
/blacklist-sync and currently removes 3 non-LLM models: grok-imagine-image, grok-imagine-image-quality, and
grok-imagine-video. Retired Grok 2/3/4 families are not listed because they no longer appear in current authoritative
catalog sources.
DigitalOcean blacklist (provider.digitalocean.blacklist)
This blacklist filters premium models that require higher account tiers on DigitalOcean's AI platform. It is
synchronized via /blacklist-sync and currently blocks 37 models: 16 anthropic-* Claude variants and 21 openai-*
models. Open-source openai-gpt-oss-* variants are preserved through the oss substring rule, and open or partner
models from Qwen, DeepSeek, Llama, Mistral, Kimi, GLM, and similar families remain available.
The oh-my-openagent.json file contains a sophisticated model assignment system that maps specialized agents to
appropriate primary models, fallback models, and optional ultrawork models for enhanced capability when needed. This
configuration represents a carefully tuned balance between API rate limits, response quality, and cost management.
Agent-Specific Model Assignments
Individual agents from the oh-my-openagent plugin receive specialized model assignments optimized for their specific functions:
| Source | Agent Name | Role | Model | Variant | Fallback Models | Description |
|---|---|---|---|---|---|---|
| oh-my-openagent | sisyphus |
Orchestrator | digitalocean/kimi-k2.6 |
— | openai/gpt-5.5 (medium), zai-coding-plan/glm-5.1, digitalocean/glm-5, opencode/big-pickle |
Primary orchestrator for complex, multi-step tasks |
| oh-my-openagent | metis |
Scope Analysis | openai/gpt-5.5 |
high |
zai-coding-plan/glm-5.2, digitalocean/kimi-k2.5 |
Pre-planning consultation and scope analysis |
| oh-my-openagent | prometheus |
Planning Specialist | openai/gpt-5.5 |
high |
zai-coding-plan/glm-5.2, github-copilot/gemini-3.1-pro-preview |
Detailed plans and work breakdowns |
| oh-my-openagent | atlas |
Knowledge Specialist | digitalocean/kimi-k2.6 |
— | openai/gpt-5.5 (medium), nvidia/minimaxai/minimax-m2.7 |
Knowledge retrieval and architectural context |
| oh-my-openagent | sisyphus-junior |
Lightweight Orchestrator | digitalocean/kimi-k2.6 |
— | openai/gpt-5.5 (medium), nvidia/minimaxai/minimax-m2.7, opencode/big-pickle |
Category-optimized task delegation |
| oh-my-openagent | hephaestus |
Implementation Specialist | openai/gpt-5.4 |
medium |
— | Executes implementation tasks with balanced capability and efficiency. Ultrawork: openai/gpt-5.5 (medium) |
| oh-my-openagent | oracle |
Strategic Advisor | openai/gpt-5.5 |
high |
github-copilot/gemini-3.1-pro-preview (high), zai-coding-plan/glm-5.2, zai-coding-plan/glm-5.1 |
Provides high-level architectural guidance and complex reasoning for critical decisions |
| oh-my-openagent | momus |
Quality Review | openai/gpt-5.5 |
xhigh |
github-copilot/gemini-3.1-pro-preview (high), zai-coding-plan/glm-5.2, zai-coding-plan/glm-5.1 |
Reviews work plans and implementations for quality, completeness, and adherence to best practices |
| oh-my-openagent | multimodal-looker |
Visual Analysis | openai/gpt-5.5 |
medium |
digitalocean/kimi-k2.6, zai-coding-plan/glm-5v-turbo |
Analyzes visual content, images, and multimodal inputs for comprehensive understanding |
| oh-my-openagent | explore |
Codebase Analysis | nvidia/minimaxai/minimax-m2.7 |
— | digitalocean/deepseek-3.2, openai/gpt-5.4-mini, openai/gpt-5.4-nano |
Performs rapid codebase navigation, pattern detection, and symbol exploration |
| oh-my-openagent | librarian |
Research Specialist | nvidia/minimaxai/minimax-m2.7 |
— | digitalocean/deepseek-3.2, openai/gpt-5.4-mini, openai/gpt-5.4-nano |
Handles documentation lookup, external research, and information retrieval tasks |
| opencode-historian | historian |
Memory Management | openai/gpt-5.4-mini |
— | — | Manages persistent memories, context retention, and semantic search across project knowledge base |
Current API Rate Limits and Suggested Setup
The provider configuration considers several factors for optimal performance:
-
Rate Limit Management: Different providers have varying rate limits. Higher-concurrency providers such as DigitalOcean, OpenCode, and Z.ai handle broader fallback coverage, while tighter limits protect heavier OpenAI, NVIDIA, and GitHub Copilot routes.
-
Cost Optimization: The configuration keeps free OpenCode models available, uses DigitalOcean open and partner models heavily, and blacklists paid Zen plus premium DigitalOcean models that should not be selected routinely.
-
Performance Tiering: Models are tiered by capability and speed:
- Flash Variants (
*-flash): Fastest responses, lowest cost, ideal for quick lookups and simple tasks - Standard Variants (no suffix): Balanced performance for general-purpose work
- High/Max Variants: Enhanced capability for complex reasoning tasks
- Flash Variants (
-
Suggested Usage Pattern:
- Use @sisyphus for complex, multistep tasks requiring coordination and agent delegation
- Use @sisyphus-junior for category-optimized task delegation via the task() system
- Use @prometheus for detailed project planning and work breakdowns
- Use @explore for codebase navigation, pattern detection, and symbol lookup
- Use @librarian for documentation research and external API lookup
- Use @oracle for high-stakes architectural decisions and complex reasoning
- Use @metis for task scope analysis and pre-planning consultation
- Use @momus for quality reviews and adherence verification
- Use @hephaestus for implementation tasks and coding workflows
- Use @atlas for knowledge retrieval and architectural context
- Use @multimodal-looker for visual content analysis and image understanding
- Use Task Categories (
visual-engineering,ultrabrain,quick,writing) for automatic model routing - Reserve High-effort Routes (
high,xhigh,max, andultrawork) for tasks where quality is critical
This configuration represents a personalized setup balancing performance, cost, and reliability based on individual usage patterns and provider strengths.
Task Category Model Assignments
The oh-my-openagent.json configuration also defines task category model assignments that automatically route tasks to
appropriate models based on their category:
| Category | Model | Variant | Fallback Models | Description |
|---|---|---|---|---|
visual-engineering |
github-copilot/gemini-3.1-pro-preview |
high |
digitalocean/glm-5, zai-coding-plan/glm-5.2, zai-coding-plan/glm-5.1, digitalocean/kimi-k2.5 |
Frontend, UI/UX, design, styling, and animation tasks |
artistry |
github-copilot/gemini-3.1-pro-preview |
high |
openai/gpt-5.5 |
Complex problem-solving with unconventional, creative approaches |
ultrabrain |
openai/gpt-5.5 |
xhigh |
github-copilot/gemini-3.1-pro-preview (high), zai-coding-plan/glm-5.2, zai-coding-plan/glm-5.1 |
Hard logic-heavy tasks requiring deep reasoning |
deep |
openai/gpt-5.5 |
medium |
github-copilot/gemini-3.1-pro-preview (high) |
Goal-oriented autonomous problem-solving with thorough research |
quick |
openai/gpt-5.4-mini |
— | github-copilot/gemini-3-flash-preview, nvidia/minimaxai/minimax-m2.7, opencode/big-pickle |
Trivial tasks, single file changes, typo fixes |
unspecified-low |
digitalocean/kimi-k2.6 |
— | openai/gpt-5.3-codex (medium), github-copilot/gemini-3-flash-preview, nvidia/minimaxai/minimax-m2.7 |
Low-effort tasks that don't fit other categories |
unspecified-high |
openai/gpt-5.5 |
high |
digitalocean/glm-5, digitalocean/kimi-k2.5 |
High-effort tasks that don't fit other categories |
writing |
digitalocean/kimi-k2.5 |
— | github-copilot/gemini-3-flash-preview, digitalocean/kimi-k2.6, nvidia/minimaxai/minimax-m2.7 |
Documentation, prose, and technical writing tasks |
git |
github-copilot/gpt-5-mini |
— | opencode/big-pickle, zai-coding-plan/glm-4.5-air |
All git operations with focus on atomic commits and safe operations |
These category assignments enable intelligent task routing, ensuring each type of work is handled by the most suitable model for optimal results.
Background Task Configuration
The oh-my-openagent.json file includes sophisticated background task management settings:
| Setting | Value | Description |
|---|---|---|
defaultConcurrency |
5 | Default number of concurrent background tasks |
staleTimeoutMs |
60000 | Timeout in milliseconds before a task is considered stale |
| Provider Concurrency | Per-provider task limits for rate limit management | |
omlx |
1 | Maximum concurrent tasks for oMLX provider |
xai |
5 | Maximum concurrent tasks for xAI provider |
nvidia |
3 | Maximum concurrent tasks for NVIDIA provider |
openai |
5 | Maximum concurrent tasks for OpenAI provider |
opencode |
10 | Maximum concurrent tasks for OpenCode provider |
zai-coding-plan |
10 | Maximum concurrent tasks for Z.ai provider |
digitalocean |
3 | Maximum concurrent tasks for DigitalOcean provider |
github-copilot |
5 | Maximum concurrent tasks for GitHub Copilot provider |
| Model Concurrency | Per-model fine-grained concurrency limits | |
openai/gpt-5.5 |
1 | Concurrency limit for OpenAI GPT-5.5 |
openai/gpt-5.4 |
2 | Concurrency limit for OpenAI GPT-5.4 |
openai/gpt-5.3-codex |
3 | Concurrency limit for OpenAI GPT-5.3-codex |
openai/gpt-5.4-mini |
5 | Concurrency limit for OpenAI GPT-5.4-mini |
openai/gpt-5.4-nano |
8 | Concurrency limit for OpenAI GPT-5.4-nano |
zai-coding-plan/glm-4.5-air |
5 | Concurrency limit for GLM-4.5-air model |
zai-coding-plan/glm-4.7 |
2 | Concurrency limit for GLM-4.7 model |
zai-coding-plan/glm-5-turbo |
1 | Concurrency limit for GLM-5-turbo model |
zai-coding-plan/glm-5v-turbo |
1 | Concurrency limit for GLM-5v-turbo model |
zai-coding-plan/glm-5.1 |
10 | Concurrency limit for GLM-5.1 model |
zai-coding-plan/glm-5.2 |
5 | Concurrency limit for GLM-5.2 model |
Runtime Fallback Configuration
Automatic fallback system for handling API errors and maintaining workflow continuity:
| Setting | Value | Description |
|---|---|---|
enabled |
true | Enable automatic fallback on errors |
max_fallback_attempts |
3 | Maximum number of fallback attempts per request |
cooldown_seconds |
60 | Cooldown period between fallback attempts |
timeout_seconds |
30 | Request timeout threshold |
notify_on_fallback |
true | Notify user when fallback occurs |
| Retry on Errors | HTTP status codes that trigger fallback | |
| 400, 429, 503, 529 | Bad Request, Rate Limited, Service Unavailable, Overloaded |
This configuration ensures robust operation by automatically switching to fallback models when primary models encounter rate limits or service issues, maintaining workflow continuity without manual intervention.
The OpenCode configuration utilizes several plugins to extend its core functionality. These plugins are defined in
opencode.json configuration file and provide integration with external services and additional features.
oh-my-openagent@latest
The oh-my-openagent plugin is a comprehensive agent collection for OpenCode that provides a full suite of specialized agents for various development tasks. This plugin delivers a robust set of agents optimized for efficient task delegation, complex problem-solving, and comprehensive development workflows. The oh-my-openagent suite includes agents for orchestration, exploration, strategic decision-making, visual engineering, research, and more, providing enterprise-grade capabilities for demanding development scenarios.
@nick-vi/opencode-type-inject@latest
The type-inject plugin provides advanced type inference and injection capabilities for OpenCode. This plugin enhances the AI's understanding of type systems across different programming languages, enabling more accurate code completion and type-aware refactoring operations. It integrates with OpenCode's language server protocol to provide real-time type information and suggestions.
opencode-openai-codex-auth@latest
The opencode-openai-codex-auth plugin provides authentication for the OpenAI provider, enabling direct API access to
OpenAI's GPT-5 family models. This plugin handles API key management and authentication flows required for the openai
provider configured in opencode.json, which is used as the primary model source for several agents including
hephaestus, oracle, momus, and multimodal-looker.
opencode-historian@latest
The historian plugin provides persistent memory management capabilities for OpenCode, enabling context retention and compounded engineering practices across sessions. This plugin allows agents to store, recall, and manage memories including architectural decisions, design patterns, learnings, preferences, issues, and contextual information. The historian system automatically classifies memory types and manages circular references between related memories, creating a knowledge base that persists beyond individual conversations.
Model Context Protocol (MCP) servers extend OpenCode's capabilities by providing specialized tools and integrations.
This repository now keeps OpenCode's local MCP surface intentionally small. OpenCode connects to one local
mcp-proxy endpoint for shared MCP servers used across OpenCode, Claude Code, Codex, and other AI agents. The one
intentional exception is Oh-My-OpenAgent's custom lsp MCP, which remains plugin-managed because it does not have a
public standalone MCPProxy configuration.
The following MCP is explicitly configured in opencode.json:
mcp-proxy
OpenCode connects to MCPProxy through the local remote endpoint http://127.0.0.1:8081/mcp. MCPProxy is the shared MCP
control plane for this workstation and exposes the actual upstream MCP servers through one OpenCode entrypoint. This
avoids duplicating MCP configuration across multiple AI coding tools.
The proxy configuration currently includes these upstream MCP servers:
| Upstream MCP | Status | Protocol | Purpose |
|---|---|---|---|
ast_grep |
Enabled | stdio | AST-aware code search and structural matching |
atlassian |
Enabled | stdio | Jira and Confluence access through mcp-atlassian |
figma |
Disabled | HTTP | Figma desktop MCP endpoint at http://127.0.0.1:3845/mcp |
github |
Enabled | HTTP | GitHub Copilot MCP remote endpoint at https://api.githubcopilot.com/mcp |
grep_app |
Enabled | HTTP | Public GitHub code search through grep.app |
serena |
Enabled | stdio | Code intelligence, symbol navigation, and semantic project operations |
vision |
Enabled | stdio | Z.ai vision MCP for image and visual-content analysis |
MCPProxy also enables management, keyword-based tool search/routing, quarantine, metrics, observability, OAuth support, sensitive-data detection, and registry discovery. The search layer acts as a real lazy-load mechanism: agents retrieve only the MCP tools needed for the current task instead of loading every available tool into the prompt. Docker isolation is configured but disabled globally in the proxy configuration.
The oh-my-openagent and opencode-historian plugin MCPs have been removed from normal local ownership and moved to
MCPProxy where possible. The retained exception is Oh-My-OpenAgent's custom lsp MCP.
| Source | Configuration File | Current Routing |
|---|---|---|
| oh-my-openagent | oh-my-openagent.json |
Use MCPProxy for moved MCPs; keep plugin lsp MCP |
| opencode-historian | opencode-historian.json |
Use MCPProxy-managed serena instead |
| Claude Code override | oh-my-openagent.json |
Avoid duplicate Claude Code plugin bridge activation |
This means the plugin agents remain available, MCPProxy is the source of truth for shared MCP server configuration, and
lsp remains the only intentionally plugin-managed OMO MCP.
Project-scoped custom slash commands live in .opencode/commands/. The current repository has two project-scoped
commands and no root-level ./commands/ directory. The make sync command will still mirror a root ./commands/
directory into the system config if one is added later.
The /blacklist-sync command dynamically synchronizes the OpenCode provider blacklists for xAI, OpenCode Zen, and
DigitalOcean. On each run it re-derives flagship status, pricing, and model membership from official sources so the
blacklist stays accurate as providers ship new models without requiring manual updates.
| Provider | Blacklist Criteria |
|---|---|
xai |
Non-LLM/multimodal models + models from outdated families (coding models always preserved) |
opencode |
Paid Zen models (free-tier models always preserved) |
digitalocean |
Premium Anthropic Claude + premium OpenAI GPT/o-series models (open-source gpt-oss-* variants always preserved) |
Source: .opencode/commands/blacklist-sync.md
The /readme-sync command synchronizes this README with the current OpenCode configuration files. It scans
opencode.json, oh-my-openagent.json, root JSON files, the Makefile, and local command or agent directories, then
updates only the README sections that drift from the actual configuration.
Source: .opencode/commands/readme-sync.md
The tui.json file contains the Terminal User Interface configuration for OpenCode:
Current Configuration:
- Theme:
opencode- Uses the default OpenCode theme for the terminal interface
This minimal configuration enables the standard OpenCode TUI experience. Additional TUI customization options can be added to this file as needed.
OpenCode employs a sophisticated agent system where specialized AI agents handle different types of tasks. The agent architecture enables parallel task execution, context-aware processing, and delegation based on task complexity and requirements.
There are no repo-local agents/ or .opencode/agents/ files in this project at the moment. The agents documented
below are provided by configured plugins.
The oh-my-openagent plugin provides a comprehensive suite of specialized agents designed for various development tasks:
Core Agents:
-
@sisyphus - The primary orchestrator that coordinates complex, multi-step tasks with heavy planning and parallel execution potential. Sisyphus manages agent delegation workflows and ensures tasks are routed to the most appropriate specialist agents.
-
@sisyphus-junior - Lightweight orchestrator that serves as the backing agent for all
task()category delegations (e.g.,visual-engineering,ultrabrain,quick). Runs on a faster model than the full Sisyphus, optimized for category-specific task execution. -
@hephaestus - Implementation specialist that executes well-defined coding tasks with efficiency and precision. Hephaestus excels at translating plans into working code and handling technical implementations.
-
@oracle - Strategic advisor that provides high-level architectural guidance and complex reasoning for high-stakes decisions. Oracle is consulted for critical architectural choices and difficult technical problems.
-
@librarian - Research specialist that handles external documentation lookup, API research, and library information retrieval. Librarian maintains up-to-date knowledge of libraries, frameworks, and best practices.
-
@explore - Codebase analysis agent that performs rapid navigation, pattern detection, and symbol exploration across the entire codebase. Explore is optimized for finding code patterns and understanding existing implementations.
-
@multimodal-looker - Visual analysis agent specialized in processing images, screenshots, diagrams, and visual content. This agent enables understanding of visual inputs alongside code and text.
Planning & Quality Agents:
-
@prometheus - Planning specialist that creates detailed work breakdowns and project plans for complex implementations. Prometheus structures tasks into manageable units with clear dependencies and execution order.
-
@metis - Scope analysis agent that analyzes task requirements, identifies ambiguities, and provides pre-planning consultation. Metis helps clarify requirements before implementation begins.
-
@momus - Quality review agent that evaluates work plans and implementations against rigorous standards. Momus ensures completeness, verifiability, and adherence to best practices.
-
@atlas - Knowledge specialist that manages and retrieves contextual information, architectural decisions, and project conventions. Atlas maintains the project's accumulated wisdom and helps agents access relevant context.
These agents work together to provide comprehensive coverage of development tasks through a sophisticated orchestration system that matches tasks to the most appropriate specialist.
The opencode-historian plugin provides a specialized agent for persistent memory management:
@historian - Memory Management Specialist
The historian agent manages persistent memories, enabling context retention and compounded engineering practices across sessions. It stores, recalls, and manages project knowledge including architectural decisions, design patterns, learnings, preferences, issues, and contextual information.
Key capabilities:
- Memory Storage: Stores project decisions, learnings, and context that persist across sessions
- Semantic Search: Retrieves relevant memories using keyword or semantic search
- Memory Classification: Automatically categorizes memories by type (architectural decisions, conventions, preferences, context)
- Cross-Reference Management: Handles circular references between related memories
The historian agent uses the openai/gpt-5.4-mini model configured in opencode-historian.json for fast,
efficient memory operations.
Comprehensive reference mapping for plugins, MCPs, external services, and related resources used in this OpenCode configuration.
| Plugin Name | Status | Official Repository | Version | Purpose |
|---|---|---|---|---|
| oh-my-openagent | Active | https://github.com/code-yeongyu/oh-my-openagent | Latest | Multi-agent orchestration suite for task delegation and complex workflows |
| @nick-vi/opencode-type-inject | Active | https://github.com/nick-vi/type-inject | Latest | Advanced type inference and language-aware code completion |
| opencode-openai-codex-auth | Active | https://github.com/numman-ali/opencode-openai-codex-auth | Latest | OpenAI provider authentication for direct API access |
| opencode-historian | Active | https://github.com/5kahoisaac/opencode-historian | Latest | Persistent memory management and semantic search across project knowledge |
| MCP Name | Status | Source Type | Source Location | Documentation | Purpose |
|---|---|---|---|---|---|
| mcp-proxy | Enabled | Local HTTP MCP | http://127.0.0.1:8081/mcp |
https://docs.mcpproxy.app/ | Shared MCPProxy endpoint for OpenCode and other tools |
| MCP Name | Status | Source Type | Documentation/Source | Purpose |
|---|---|---|---|---|
| ast_grep | Enabled | Git repository | https://github.com/ast-grep/ast-grep-mcp | AST-aware code search and structural matching |
| atlassian | Enabled | Python package | https://github.com/sooperset/mcp-atlassian | Atlassian Jira and Confluence integration |
| figma | Disabled | Local HTTP endpoint | https://help.figma.com/hc/en-us/articles/32132100833559-Guide-to-the-Figma-MCP-server | Figma design integration and UI code generation |
| github | Enabled | Remote endpoint | https://api.githubcopilot.com/mcp | GitHub repository operations, pull requests, issues, and code search |
| grep_app | Enabled | Remote endpoint | https://vercel.com/blog/grep-a-million-github-repositories-via-mcp | GitHub code search through grep.app |
| serena | Enabled | Git repository | https://github.com/oraios/serena | Code intelligence, symbol navigation, and semantic project operations |
| vision | Enabled | npm package | https://docs.z.ai/devpack/mcp/vision-mcp-server | Z.ai vision MCP for image understanding and visual analysis |
| MCP Name | Status | Source | Documentation/Source | Purpose |
|---|---|---|---|---|
| lsp | Enabled | oh-my-openagent | Internal plugin MCP | OMO custom LSP MCP retained because no public standalone proxy config exists |
| Service | URL | Purpose |
|---|---|---|
| OpenCode Platform | https://opencode.ai | Main OpenCode AI coding assistant platform |
| OpenCode Models | https://opencode.ai/docs/models | Official reference for model configuration and provider selection |
| OpenCode Zen | https://opencode.ai/docs/zen | Official Zen model hub and pricing roster documentation |
| GitHub Copilot Docs | https://docs.github.com/en/copilot | Official GitHub Copilot setup and usage documentation |
| GitHub Models Catalog | https://docs.github.com/en/rest/models/catalog | Canonical API reference for GitHub-hosted model catalog metadata |
| NVIDIA NIM Docs | https://docs.nvidia.com/nim/large-language-models/latest/introduction.html | Official NVIDIA NIM documentation for hosted LLM access and model serving |
| OpenAI Platform | https://developers.openai.com/api/docs/models | Official OpenAI model documentation for GPT-5 family and API reference |
| Kimi API Platform | https://platform.kimi.ai/docs/models | Official model documentation for Kimi K2.6, K2.5, and related families |
| Z.ai Developer Docs | https://docs.z.ai/guides/overview/quick-start | Official Z.ai documentation for GLM model families and APIs |
| xAI Developer Docs | https://docs.x.ai/developers/models | Official xAI model catalog and pricing overview |
| Model Context Protocol | https://modelcontextprotocol.io/docs/getting-started/intro | Open standard for AI-LLM context and tool integration (Anthropic) |
| MCPProxy Docs | https://docs.mcpproxy.app/ | Official MCPProxy documentation for shared MCP server management and tool search |
| MCPProxy Routing Modes | https://docs.mcpproxy.app/features/routing-modes/ | retrieve_tools routing mode docs for token-saving lazy tool retrieval |
| MCPProxy Config Gist | https://gist.github.com/5kahoisaac/319e1eb499fe5902a93dbe3d7cb0bf1a | User-maintained MCPProxy upstream configuration reference |
- MCPProxy:
opencode.jsonpoints only at the local MCPProxy endpoint. Upstream MCP details live in the proxy configuration, not in OpenCode's local config. - MCPProxy Config Gist: Reference link only; secret values should stay managed by keyring or local MCPProxy config, not copied into this README.
- Figma MCP: No public implementation repository exists; reference is official Figma documentation only.
- GitHub MCP: MCPProxy points at the remote GitHub Copilot MCP endpoint
https://api.githubcopilot.com/mcp. - Atlassian MCP: Community fork maintained by sooperset; official Atlassian server exists but uses remote HTTP.
- Vision MCP: Connects through Z.ai's MCP package for image analysis.
- OMO
lspMCP: Retained as a plugin-managed MCP because it is custom to Oh-My-OpenAgent and has no public standalone MCPProxy configuration.
All URLs verified as of June 2026. Refer to individual repository documentation for latest API changes and version compatibility.