docs: add multi-agent setup guide #258
Conversation
|
Note Other AI code review bot(s) detectedCodeRabbit has detected other AI code review bot(s) in this pull request and will avoid duplicating their findings in the review comments. This may lead to a less comprehensive review. WalkthroughAdds a new documentation file at docs/MULTI-AGENT-SETUP.md describing a multi-agent architecture: a Primary Agent using Antigravity for complex reasoning, Fast subagents using Cerebras for simple/low-latency tasks, and an optional security-auditor agent. Defines model tiers (Heavy and Fast), provides example configurations for primary, subagent, and auditor roles, and discusses rationale for rate-limit management and balancing quality vs. cost. Related links are included. Estimated code review effort🎯 1 (Trivial) | ⏱️ ~4 minutes 🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. 📜 Recent review detailsConfiguration used: Repository UI Review profile: CHILL Plan: Pro Cache: Disabled due to data retention organization setting Knowledge base: Disabled due to data retention organization setting 📒 Files selected for processing (1)
🚧 Files skipped from review as they are similar to previous changes (1)
⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (1)
✏️ Tip: You can disable this entire section by setting Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
Greptile SummaryAdds comprehensive documentation for multi-agent configuration that helps users reduce Antigravity rate limit usage by delegating simple tasks to free providers like Cerebras while reserving Antigravity for complex reasoning tasks.
Confidence Score: 5/5
Important Files Changed
Sequence DiagramsequenceDiagram
participant User
participant Primary as Primary Agent<br/>(Antigravity Claude Opus)
participant Explore as Explore Subagent<br/>(Cerebras llama-3.3-70b)
participant Researcher as Researcher Subagent<br/>(Cerebras llama-3.3-70b)
participant Coder as Coder Subagent<br/>(Cerebras llama-3.3-70b)
participant Security as Security Auditor<br/>(Antigravity Sonnet)
participant Antigravity as Antigravity API<br/>(Rate Limited)
participant Cerebras as Cerebras API<br/>(Free, Fast)
User->>Primary: Complex task request
Note over Primary: Complex reasoning<br/>uses Antigravity quota
Primary->>Antigravity: API call (quota used)
Antigravity-->>Primary: Response
Primary->>Explore: Delegate exploration task
Explore->>Cerebras: API call (free, fast)
Cerebras-->>Explore: Response (~2100 tok/sec)
Explore-->>Primary: Exploration results
Primary->>Researcher: Delegate research task
Researcher->>Cerebras: API call (free, fast)
Cerebras-->>Researcher: Response
Researcher-->>Primary: Research findings
Primary->>Coder: Delegate code generation
Coder->>Cerebras: API call (free, fast)
Cerebras-->>Coder: Response
Coder-->>Primary: Generated code
Primary->>Security: Security review (needs thinking)
Security->>Antigravity: API call (quota used)
Antigravity-->>Security: Security analysis
Security-->>Primary: Security report
Primary->>User: Complete response
Note over Primary,Cerebras: Result: Reduced Antigravity quota usage<br/>Most API calls go to free Cerebras<br/>Complex reasoning still uses Claude thinking
|
There was a problem hiding this comment.
Pull request overview
This PR adds a new documentation guide explaining how to set up a multi-agent architecture to reduce rate limit pressure by using Antigravity only for complex tasks and free providers (like Cerebras) for faster subagent operations.
Changes:
- Added
docs/MULTI-AGENT-SETUP.mdwith architecture diagram, model tier recommendations, and example agent configurations - Documented how to balance between Antigravity's premium models and free third-party providers for optimal rate limit management
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
docs/MULTI-AGENT-SETUP.md
Outdated
| ```markdown | ||
| <!-- ~/.config/opencode/agent/security-auditor.md --> | ||
| --- | ||
| model: google/antigravity-claude-sonnet-4-5 |
There was a problem hiding this comment.
The section header states "needs thinking" but the model configuration uses google/antigravity-claude-sonnet-4-5 which is the non-thinking variant. According to the README, antigravity-claude-sonnet-4-5-thinking is the model with extended thinking capabilities. Either update the model to use the -thinking variant, or remove the "needs thinking" comment from the header to match the actual model being used.
| model: google/antigravity-claude-sonnet-4-5 | |
| model: google/antigravity-claude-sonnet-4-5-thinking |
There was a problem hiding this comment.
Fixed - added -thinking suffix to the security auditor model.
docs/MULTI-AGENT-SETUP.md
Outdated
|
|
||
| | Tier | Provider | Models | Use Case | | ||
| |------|----------|--------|----------| | ||
| | Heavy | Antigravity | claude-opus-4-5-thinking, claude-sonnet-4-5-thinking | Complex reasoning, architecture | |
There was a problem hiding this comment.
The model names in the table lack the antigravity- prefix that is required for Antigravity models. Based on the configurations shown in the examples and the README, these should be antigravity-claude-opus-4-5-thinking and antigravity-claude-sonnet-4-5-thinking to be consistent with actual model identifiers used in the codebase.
| | Heavy | Antigravity | claude-opus-4-5-thinking, claude-sonnet-4-5-thinking | Complex reasoning, architecture | | |
| | Heavy | Antigravity | antigravity-claude-opus-4-5-thinking, antigravity-claude-sonnet-4-5-thinking | Complex reasoning, architecture | |
There was a problem hiding this comment.
Fixed - added antigravity- prefix to model names in the table.
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 3
🤖 Fix all issues with AI agents
In `@docs/MULTI-AGENT-SETUP.md`:
- Around line 7-13: Add a language specifier to the markdown code fence around
the architecture diagram by changing the opening triple backticks to include a
language (e.g., ```text); update the code block that starts with "Primary Agent
(Antigravity Claude Opus)" so the block is fenced as ```text (or another
appropriate language) to satisfy markdownlint and ensure correct rendering.
- Around line 15-21: Update the Security Auditor agent's model string to use the
verified "thinking" variant: replace the current model value
"google/antigravity-claude-sonnet-4-5" with
"google/antigravity-claude-sonnet-4-5-thinking" in the Security Auditor agent
configuration (the model: field for the Security Auditor entry) so it matches
the Model Tiers table and ANTIGRAVITY_API_SPEC.md.
- Around line 73-76: The "OpenCode Agent Docs" link currently points to a 404
URL; update the markdown entry that references "OpenCode Agent Docs" (the link
text with URL https://opencode.ai/docs/customization/agents) to the correct
OpenCode agents customization documentation URL by replacing the broken href
with the valid docs URL (verify the correct destination on opencode.ai or GitHub
and update the link target accordingly), ensuring the adjacent GitHub issue
reference [OpenCode `#7138`] remains unchanged.
📜 Review details
Configuration used: Repository UI
Review profile: CHILL
Plan: Pro
Cache: Disabled due to data retention organization setting
Knowledge base: Disabled due to data retention organization setting
📒 Files selected for processing (1)
docs/MULTI-AGENT-SETUP.md
🧰 Additional context used
🪛 markdownlint-cli2 (0.18.1)
docs/MULTI-AGENT-SETUP.md
7-7: Fenced code blocks should have a language specified
(MD040, fenced-code-language)
⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (2)
- GitHub Check: Upload results
- GitHub Check: Greptile Review
🔇 Additional comments (4)
docs/MULTI-AGENT-SETUP.md (4)
1-4: LGTM! Clear and concise introduction.The title and introduction effectively communicate the guide's purpose and align with the PR objectives.
22-34: LGTM! Clear primary agent configuration.The example demonstrates the
mode: primaryconfiguration with Antigravity Claude Opus as specified in the PR objectives.
36-49: LGTM! Excellent subagent configuration.The example properly demonstrates
mode: subagentwith Cerebras as a free provider, and appropriately restricts tools to read-only operations for exploration tasks.
66-72: Correct the Cerebras throughput specification.Line 69 claims Cerebras provides "8000+ tokens/sec" throughput, but current specs show Cerebras Llama 3.3 (70B) actually delivers approximately 2,100–2,600 tokens/second. The "free" claim is accurate; update the speed figure to reflect actual performance.
✏️ Tip: You can disable this entire section by setting review_details to false in your review settings.
chindris-mihai-alexandru
force-pushed
the
docs/multi-agent-setup
branch
from
3b05339 to
b2f7517
Compare
|
Thanks for the contribution! This is outside the scope of this plugin—we focus on the Antigravity auth layer, not general OpenCode configuration patterns. Feel free to post this idea in the Discussions section where it might get more visibility from the community. |
ok, I see. will do! |
2 participants
Adds
docs/MULTI-AGENT-SETUP.mdshowing how to reduce rate limits by using Antigravity only for complex tasks and free providers (like Cerebras) for fast subagents.Closes #257