OpenConfig — Best OpenCode Config · oh-my-openagent · OpenRouter · AI Coding Agent

THE GREATEST CODING AGENT CONFIGURATION EVER ASSEMBLED

OpenConfig · OpenCode · oh-my-openagent (OmO) · OpenRouter · Claude · GPT · DeepSeek · Gemini · GLM Exacto · MCP · Context7 · Exa · tmux · Ghostty

v1.5.22 · CLI oc · identity openconfig/opencode-configs

GitHub topics / search keywords: opencode · oh-my-openagent · oh-my-opencode · openrouter · coding-agent · ai-agent · llm · claude · openai · gpt · deepseek · gemini · glm · agentic · mcp · context7 · exa · tmux · ghostty · developer-tools · cli · sisyphus · hephaestus · hyperplan · ultrawork · ralph-loop · multi-agent · ai-coding · opencode-config · best-opencode-config

Not a theme. Not a “dotfiles dump.” Not vibes.

This is the best-known, battle-hardened, over-tuned, footgun-exterminated global stack for OpenCode + OpenRouter + oh-my-openagent (OmO) — the configuration that turns a raw coding CLI into a multi-model AI coding agent war machine with self-healing, live doctor checks, adversarial planning, content-aware research, and model routing so sharp it feels unfair.

If other OpenCode / OmO / OpenRouter setups are “pretty good,” OpenConfig is the final form.

# Clone (or already at ~/.config/opencode)
git clone https://github.com/jesseoue/opencode-configs.git
cd opencode-configs
oc install --quick          # or: ./install.sh --yes

# Fresh machine (no clone yet)
curl -fsSL https://raw.githubusercontent.com/jesseoue/opencode-configs/main/install.sh | bash
source ~/.zshrc && oc doctor && oc launch


Pinned excellence	OpenConfig `1.5.22` · OpenCode `1.18.4+` · OmO `[email protected]`
Default god-mode lead	`sisyphus` (GLM Exacto — tool-call royalty)
Config path	`~/.config/opencode` → this repo (symlink)
Projects home	`oc new` → `~/Projects/<name>`
Verdict	`oc doctor` → Ready to code — everything checks out.

Plugin name must stay oh-my-openagent@… (not legacy oh-my-opencode).
Schema URL basename stays oh-my-opencode.schema.json (upstream asset; the oh-my-openagent.schema.json path 404s).

Why OpenConfig is the best OpenCode + oh-my-openagent + OpenRouter coding agent config

Most “AI agent setups” are a random JSON file, three conflicting prompts, and a prayer.

OpenConfig is the opposite:

Pinned models that actually win their jobs — Exacto for orchestration, Nitro for parallel recon, direct OpenAI Sol for deep implement/critique, Gemini for visual/writing, Claude Fable for the hard ceiling.
A full agent pantheon — Sisyphus leads. Hephaestus ships. Oracle adjudicates. Explore maps. Librarian docs. Content-aware research goes nuclear without soft refusals.
Team mode that doesn’t melt your wallet — intentional concurrency caps, hyperplan adversarial planning, Goal kept off (OmO 4.19 /start-work footgun).
Research stack that refuses to hallucinate APIs — local → Context7 → grep_app → Exa → webfetch. In that order. Encoded in prompts. Enforced by culture.
Self-healing ops — oc validate · oc doctor · oc heal · oc signature · oc test. If it drifts, it tells you. If it’s broken, it repairs itself.
Telemetry dark. Secrets local. Identity proven. This tree fingerprints as OpenConfig — not a random clone of someone else’s weekend experiment.

Decision log: AGENTS.md · Day-to-day stance: prompts/core.md · Changelog: CHANGELOG.md

Quick start — install OpenConfig OpenCode oh-my-openagent OpenRouter in 60 seconds

# Seed keys (required / recommended)
export OPENROUTER_API_KEY=…     # required — the bloodstream
export OPENAI_API_KEY=…         # GPT lane (Hephaestus / Oracle / Momus / …)
export EXA_API_KEY=…            # websearch that doesn’t suck
export CONTEXT7_API_KEY=…       # library docs that are actually true

oc install --quick
oc signature && oc test && oc doctor
oc launch

Or edit keys after install:

$EDITOR ~/.config/opencode/.env   # chmod 600; never commit
source ~/.zshrc
oc doctor && oc launch

One install. Paste keys. Doctor green. Code like you have unfair advantages — because you do.

`oc` CLI commands — OpenConfig doctor validate heal launch for OpenCode

oc install --quick     # install / refresh the entire stack
oc check               # validate + doctor --quick (health in one shot)
oc heal                # probe-first self-repair (it fixes itself)
oc launch [dir]        # TUI (never starts in the config repo)
oc new myapp           # scaffold under ~/Projects
oc run "…"             # headless to completion
oc admin health        # live OpenRouter + OpenAI probes
oc locate              # where is repo / CLI / keys?
oc signature           # prove this tree is OpenConfig
oc test                # smoke + idempotency sandbox
oc doctor              # full readiness — concurrency, teams, MCP, the works

Prefer oc <cmd> over raw ./foo.sh. Full help: oc help.

Research stack — Context7 · Exa · grep_app · MCP · codegraph · LSP for AI coding

Need	Tool	Why it’s elite
Local code	`read` · `grep` · `glob` · codegraph · LSP	Always first. No excuse.
Library / framework APIs	Context7	Versioned docs. Stop inventing APIs.
GitHub call sites	grep_app	Real code in the wild.
Current web / news / companies / people	Exa websearch	Ideal-page queries. Category filters.
Known URL → markdown	webfetch	Finish the job.

Exa tips: describe the ideal page, not keywords. Optional: category:company · category:people · category:news · category:research paper.

Prompts encode this matrix in prompts/core.md, sisyphus, and librarian. This is how a coding agent stops lying and starts knowing.

OpenCode agents — Sisyphus Hephaestus Oracle Explore Librarian Claude GPT DeepSeek

Primary tab agents (sisyphus · hephaestus · prometheus · atlas · content-aware-research)

Agent	Model	Role
sisyphus	GLM 5.2 Exacto	Default orchestrator / lead — owns the outcome
hephaestus	GPT-5.6 Sol (direct OpenAI)	Implementation bursts that actually land
prometheus	GLM 5.2 Exacto	Planner with teeth
atlas	GLM 5.2 Exacto	Coordination / follow-through
content-aware-research	DeepSeek V4 Pro	Full-depth research, edit denied, no soft refusals

OmO subagents (oracle · librarian · explore · momus · metis · multimodal)

Agent	Model	Role
oracle	GPT-5.6 Sol	Deep critique / adjudication
librarian	DeepSeek Flash Nitro	Docs (Context7-first)
explore	DeepSeek Flash Nitro	Codebase map at warp speed
multimodal-looker	Claude Sonnet 5	Vision
metis	Claude Sonnet 5	Strategy
momus	GPT-5.6 Sol max	Harsh review (your ego will survive)
sisyphus-junior	DeepSeek Flash Nitro	Cheap delegated work

Native OpenCode build is disabled. plan stays demoted for hyperplan handoff — do not put it in disabled_agents.

OmO categories — bug-hunt · refactor · arch-review · visual-engineering · ultrabrain

Category	Model	Use
`bug-hunt`	GLM Exacto	Reproduce → root cause → fix
`refactor-safe`	GLM Exacto	Behavior-preserving refactors
`arch-review`	GPT-5.6 Sol	Coupling / blast radius
`content-aware-fast`	DeepSeek Flash Nitro	Attack-surface recon
`content-aware-deep`	DeepSeek Pro Exacto	Deep vuln research
`writing`	Gemini 3.5 Flash Nitro	Docs / prose
`visual-engineering`	Gemini 3.1 Pro	Ship UI
`artistry`	Gemini 3.1 Pro	Design direction
`quick`	DeepSeek Flash Nitro	Cheap fast tasks
`deep` / `ultrabrain`	GPT-5.6 Sol	Heavy / max reasoning
`unspecified-low` / `unspecified-high`	Flash / Claude Fable	Hyperplan critics

Keyword triggers — ultrawork · hyperplan · team · /goal · /start-work

Say	Effect
`ultrawork` / `ulw`	Claude Fable max ceiling
`team`	Team-mode expansion
`hyperplan` / `hpp` / `/hyperplan`	Adversarial planning (from sisyphus)
`/goal`	Disabled — OmO 4.19.0 goal hook breaks `/start-work` (`InvalidObjectiveError`). Use `/start-work` → Atlas (`prompts/goal.md`)
`/start-work`	Atlas executes an approved Prometheus plan (preferred over `/goal`)

Multi-agent teams — OmO team mode · hyperplan · ship-feature · debug-team

Lead: sisyphus. Specs in teams/ are symlinked to ~/.omo/teams/ by oc setup (never leave directory copies).

Eligible members: sisyphus, atlas, sisyphus-junior, hephaestus (teammate: allow), or kind: category.
Hard-rejected as teammates: explore · librarian · oracle · metis · momus · multimodal · prometheus.

Team mode knobs (OmO 4.19): max_parallel_members=4 · max_members=5 · mailbox poll 1000ms · tmux main-vertical / inline.

Team	Members
`explorers`	deep + quick
`ship-feature`	hephaestus, junior, bug-hunt
`debug-team`	bug-hunt + ultrabrain
`review-panel`	arch-review, bug-hunt, refactor-safe
`refactor-team`	arch-review, refactor-safe
`docs-team`	writing × 2
`content-aware-audit`	content-aware-fast + content-aware-deep

Model routing — OpenRouter Exacto Nitro · OpenAI GPT · Claude · DeepSeek · Gemini · GLM

Lane	Models	Used for
Orchestration	`z-ai/glm-5.2:exacto`	Sisyphus / Atlas / Prometheus / bug-hunt
GPT (direct)	`openai/gpt-5.6-sol` (+ terra / 5.5 fallbacks)	Hephaestus / Oracle / Momus / deep
Recon	`deepseek/deepseek-v4-flash:nitro`	explore / librarian / junior / quick
Depth	`deepseek/deepseek-v4-pro:exacto`	content-aware / hard fallback
Visual / writing	Gemini 3.1 Pro · 3.6 Flash Nitro	artistry / visual / writing
Ceiling	`anthropic/claude-fable-5`	ultrawork · unspecified-high

OpenRouter is primary. GPT agents prefer direct OpenAI. Fallbacks + runtime_fallback on API errors. Stream timeouts: 900s — because real work is long.

Concurrency caps — background_task · providerConcurrency · team parallel · Goal off

Priority: modelConcurrency → providerConcurrency → defaultConcurrency. oc heal / fix.sh re-apply the caps if something drifts high.

Knob	Value	Why
`background_task.defaultConcurrency`	4	Global fan-out ceiling
OpenRouter / OpenAI / Anthropic	6 / 4 / 2	Provider budgets that don’t melt
Flash / Exacto / Sol / Fable	4 / 3 / 3 / 1	Cheap recon parallel; expensive models serial
Team `max_parallel_members` / `max_members`	4 / 5	Hyperplan floor without runaway
Goal / stale / TTL	off / 180s / 30m	Goal disabled (OmO 4.19 `/start-work` footgun); hung-task cleanup

This is what “maximum power with adult supervision” looks like.

MCP servers — Context7 docs · Exa websearch · grep_app · codegraph · TypeScript Python Go LSP

Name	Auth	Purpose
Context7	`CONTEXT7_API_KEY`	Library docs (`opencode.json` + OmO)
websearch (Exa)	`EXA_API_KEY`	Live web
grep_app	—	GitHub code search
codegraph	—	Local graph at `~/.omo/codegraph`
LSP	—	TypeScript · Python · Go only

Formatters: Prettier + Ruff. Skills fenced to ./skills (empty by design). Third-party MCPs stay disabled. Telemetry dark (OpenCode share/OTel · OmO PostHog · codegraph · DO_NOT_TRACK).

API keys — OPENROUTER_API_KEY · OPENAI_API_KEY · EXA_API_KEY · CONTEXT7_API_KEY

Key	Required	Enables
`OPENROUTER_API_KEY`	yes	OpenRouter models
`OPENAI_API_KEY`	yes for GPT lane	Hephaestus / Oracle / Momus / deep / …
`EXA_API_KEY`	for websearch	OmO Exa
`CONTEXT7_API_KEY`	recommended	Context7
`OPENROUTER_MGMT_KEY`	optional	`oc admin`
`OC_PROJECTS_DIR`	optional	`oc new` home (default `~/Projects`)

Copy .env.example → .env (chmod 600). Never commit .env.
oc setup --sync-env imports allowlisted keys only from Infisical/Doppler — never a full vault dump.

Profiles & scaffolding — oc new · high · research · content-aware · debug · writing

oc new myapp                     # ~/Projects/myapp · profile high
oc new myapp --profile research
oc new myapp --profile content-aware
oc projects --list

Profile	Agent	Use
`high`	sisyphus	Default — Exacto excellence
`low` / `fast`	sisyphus / hephaestus	Cheap / direct coding
`research` / `debug`	sisyphus	Deep reasoning / debug
`writing`	sisyphus	Documentation
`content-aware`	content-aware-research	Full-depth research (edit deny)

Each project gets opencode.json + AGENTS.md. Do not set OPENCODE_CONFIG to .opencode/profile.json.

Safety — trusted local allow · catastrophic bash deny · OpenRouter + OpenAI only

Allow-everything locally for normal tools (trusted box).
Hard-deny bash: rm -rf /|~, mkfs, sudo, git push --force*, gh repo delete*.
Providers allowed: OpenRouter + OpenAI only.
Server: 127.0.0.1:4097 · share disabled · mdns off.

Terminal cockpit — Ghostty · tmux · zsh · OpenCode TUI · Ctrl+X · Ctrl+B

Ghostty ≥ 1.3.0 · tmux ≥ 3.3 (rec. 3.7+) · zsh snippet
OpenCode leader Ctrl+X · tmux prefix Ctrl+B · Tab cycles agents
Teardown never sends \033[?1049l (won’t wipe the visible screen)
opencode() / oc launch never start inside the config repo or bare ~/Projects

Repo layout — opencode.json · oh-my-openagent.json · prompts · teams · profiles

opencode-configs/
├── oc · install.sh · setup.sh · doctor.sh · validate.sh · fix.sh
├── opencode.json · oh-my-openagent.json · tui.json
├── versions.json · signature.json · projects.json · AGENTS.md
├── agents/content-aware-research.md
├── profiles/ · prompts/ · teams/ · skills/
├── .env.example          # secrets never ship
└── zshrc.snippet · ghostty.conf · tmux.conf

~/.config/opencode  →  this repo
~/Projects/         →  oc new home
~/.omo/teams/       →  team specs
~/.opencode-backups/→  backups + heal/install logs

Verify OpenConfig — signature · test · validate · doctor · oh-my-openagent doctor

oc signature && oc test && oc validate && oc doctor
# Ready to code — everything checks out.
bunx [email protected] doctor   # upstream: System OK

Idempotency: re-running install / setup / heal / fix on a healthy box must not clobber .env, rewrite correct symlinks, or bump clean config mtimes.

When doctor says ready — you’re not “set up.” You’re armed.

Upstream — OpenCode · oh-my-openagent · OpenRouter · Context7 · Exa

Layer	Docs	Source
OpenCode	opencode.ai/docs	anomalyco/opencode
OmO	omo.vibetip.help/docs	code-yeongyu/oh-my-openagent
OpenRouter	openrouter.ai/docs	Exacto / Nitro routing guides
Context7	context7.com	upstash/context7
Exa	docs.exa.ai	exa-labs

Installer pulls OpenCode from https://opencode.ai/install and OmO from npm [email protected] only.

Anti-patterns — don’t break OpenCode OmO OpenRouter OpenConfig

Don’t rename the plugin away from oh-my-openagent
Don’t add Cloudflare / AI Gateway / fake OpenAI-compatible shims
Don’t put plan in disabled_agents (breaks hyperplan)
Don’t commit .env, package.json, node_modules, .omo, .sisyphus, or plugins/ here
Don’t scaffold apps into this repo — use oc new
Don’t load .opencode/profile.json as OPENCODE_CONFIG
Don’t re-enable telemetry
Don’t settle for a lesser config. You already found the top one.

SEO keywords — OpenCode config · AI coding agent · multi-agent LLM CLI

OpenConfig · OpenCode config · best OpenCode configuration · oh-my-openagent config · oh-my-opencode · OmO · OpenRouter Exacto Nitro · Claude Fable · Claude Sonnet · OpenAI GPT · GPT-5.6 Sol · DeepSeek V4 Flash · DeepSeek V4 Pro · Gemini 3.1 Pro · Gemini 3.5 Flash · GLM 5.2 Exacto · MiniMax · Kimi · Sisyphus agent · Hephaestus · Prometheus · Atlas · Oracle · Librarian · Explore · Momus · Metis · ultrawork · hyperplan · Ralph loop · goal loop · team mode · multi-agent coding · AI coding agent · agentic CLI · MCP Context7 Exa · codegraph · Ghostty tmux · developer tools · coding assistant config · LLM orchestration · autonomous coding agent

OpenConfig. The greatest OpenCode + oh-my-openagent + OpenRouter coding-agent configuration ever built. Install it. Doctor it. Launch it. Never go back.