Skip to main content

​
OpenCode Plugin

opencode-alchemyst-context gives opencode agents persistent, retrievable memory backed by Alchemyst AI. Agents can search context, save decisions, manage conversation memory, and ask grounded questions — all through a set of dedicated tools.

​
Features

  • Context search — retrieve relevant documents before answering
  • Context add — save decisions, snippets, and conventions
  • Memory persist — store conversation turns keyed by session
  • Grounded Q&A — ask questions and get synthesized answers
  • Security guard — heuristic outbound content blocking for secrets

​
Installation

npm install opencode-alchemyst-context
Add to opencode.json: 
{
  "plugin": ["opencode-alchemyst-context"]
}

​
Setup

On first use, tell the agent to configure the API key: 
alchemyst_configure: set my key to <your-jwt-bearer-token>
This saves to ~/.config/opencode-alchemyst/config.json and persists across sessions. You can also pre-configure manually: 
{
  "apiKey": "your-jwt-bearer-token",
  "groupName": ["opencode", "my-project"]
}

​
Environment variables

Env varDefaultPriority
ALCHEMYST_BASE_URLhttps://platform-backend.getalchemystai.comoverrides config
ALCHEMYST_DEFAULT_SCOPEinternaloverrides config
ALCHEMYST_GROUP_NAMEopencodeoverrides config

​
Tools

​
alchemyst_configure

Set the Alchemyst API key and connection settings. Required before using any other tool. Search stored context for relevant documents. Use when you need to check if decisions or reference material already exist.

​
alchemyst_context_add

Save a decision, snippet, or convention. Runs through a security guard that blocks keys, tokens, and passwords before sending.

​
alchemyst_memory_add

Persist conversation turns keyed by session ID.

​
alchemyst_memory_update

Update previously saved memory entries for the current session.

​
alchemyst_context_ask

Ask a grounded question and get a synthesized answer from stored context. 
query: "What's our policy on dependency versions?"
steeringPrompt: "answer in one sentence"

​
Security

The plugin includes an outbound content guard that blocks:
  • Private key headers (BEGIN RSA PRIVATE KEY)
  • AWS access keys (AKIA...)
  • OpenAI-style API keys (sk-...)
  • GitHub tokens (ghp_..., gho_..., ghu_...)
  • Generic SECRET/TOKEN/PASSWORD assignments
Matched content is blocked before any network request. The API key is stored with 0600 permissions and never logged.

​
Resources