evilayman

evilayman / opencode-raven

Public

Hard tool/MCP rerouter for opencode that prevents flooding and wasted context from searches, docs, web, GitHub examples, and verbose MCP calls.

ai-agent context-engineering context-management context7 exa
22
0
80% credibility
Found May 31, 2026 at 22 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
TypeScript
AI Summary

Opencode-Raven is a plugin for the opencode.ai AI coding assistant that intercepts certain tool requests (search, documentation lookup, web browsing) and routes them through a dedicated helper agent called Raven. Instead of letting raw search results flood your conversation, Raven does the work and returns a compact answer. This keeps your main assistant fast and focused, uses less of your AI budget, and gives you control over which tools get managed. Raven auto-configures on install with defaults for common tools, and you can customize routing using simple commands like '/raven route tool add' to add any tool or service prefix you want managed.

How It Works

1
💭 Your AI coding assistant gets overloaded

You notice that when you ask your assistant to search docs, browse websites, or look up examples, the conversation fills up with long, noisy results that slow everything down.

2
📦 You add Raven to your assistant

You enable the Raven plugin in your assistant's settings. Raven quietly starts running in the background, ready to help manage your tool requests.

3
⚡ Your searches still work, but stay clean

When you ask for help finding something in documentation or on the web, Raven does the searching but hands you only a clean, summarized answer instead of dumping raw results into your chat.

4
🔧 You adjust what Raven handles

You can tell Raven which tools to manage using simple commands like '/raven route tool add' to add a tool, or '/raven route mcp add' to route all tools from a particular service.

5
You talk directly to Raven when you want
📊
You check how much context you've saved

Use '/raven stats' to see how many tokens Raven has saved you across this session and all time — you might be surprised how much room you've gained.

⚙️
You fine-tune Raven's behavior

Adjust how deeply Raven thinks with '/raven effort', change how long it waits with '/raven timeout', or swap in a different AI model with '/raven model'.

✨ Everything stays fast and focused

Your main assistant stays lean and fast because Raven handles all the heavy tool work quietly in the background, returning only the answers you actually need.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 22 to 22 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is opencode-raven?

opencode-raven is a plugin for opencode that intercepts expensive tool calls and MCP requests, routing them through a dedicated "Raven" agent instead of letting them flood your main model's context. When you run a search, fetch docs, hit GitHub examples, or use verbose MCP tools, Raven catches those calls, processes them in a separate session with a cheaper model, and returns only a compact answer. The main agent never sees the raw tool output. It ships as a TypeScript package with a `/raven` command suite to toggle routing, manage which tools and MCP servers get intercepted, and view context savings stats. By default it blocks grep, glob, webfetch, fetch, and bash for non-Raven agents and routes them through `raven_seek`.

Why is it gaining traction?

Context window limits are a real pain point for anyone doing heavy agentic work. Developers are tired of watching their main model burn tokens on raw search results and documentation dumps. Raven addresses this with hard enforcement: configured tools simply fail for non-Raven agents with a direct instruction to use `raven_seek` instead of soft suggestions. The bundled MCPs (Context7, Exa, Grep.app) work without API keys out of the box. You can swap in a cheaper model for Raven's work while your main model stays focused on decisions and edits. The plugin also tracks exactly how many tokens you've saved, so you can see the impact in real time.

Who should use this?

Developers using opencode for complex, multi-tool tasks where search, docs lookups, and MCP calls eat up most of the context budget. If you're working with large codebases and find your sessions blowing up with raw tool output, this is for you. Power users who want to run a cheap model for discovery work while reserving expensive context for synthesis will get the most value. Teams with long-running agentic workflows where context efficiency directly impacts cost or quality.

Verdict

Raven solves a legitimate problem and implements it correctly, but the 22-star count means you're an early adopter shouldering some risk. The documentation is solid and the CLI is well-designed, but test coverage is unknown and the project is small. Try it if context bloat is killing your opencode sessions -- just don't bet production workflows on it yet.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.