Skip to main content
The base runtime (@lucid-fdn/agent-runtime) is a minimal Express server designed for Path B (no-code) agents. You can extend it by customizing environment variables, adding tools, enabling memory, and connecting to external services.

​
Base Runtime Architecture

The runtime runs on port 3100 and provides:
  • OpenAI-compatible chat endpoint
  • Inference routing to any provider via PROVIDER_URL
  • Automatic receipt creation via LUCID_API_URL (fire-and-forget, decoupled from inference)
User request -> Base Runtime (:3100)
  -> Forward to PROVIDER_URL (TrustGate, Ollama, vLLM, OpenAI)
  <- Response
  -> Create receipt via LUCID_API_URL (async, non-blocking)

​
Adding Tools

Enable MCP tools for your agent by setting the LUCID_TOOLS environment variable: 
lucid launch --runtime base \
  --model openai/gpt-4o \
  --prompt "You are a research assistant with web access" \
  --target docker \
  --env LUCID_TOOLS=web-search,github
Tools are accessed through MCPGate. The runtime calls MCPGate to execute tool invocations and returns results to the LLM.

​
Enabling Memory

Give your agent persistent memory across conversations: 
lucid launch --runtime base \
  --model openai/gpt-4o \
  --prompt "You are a personal assistant that remembers context" \
  --target docker \
  --env MEMORY_ENABLED=true \
  --env MEMORY_STORE=sqlite \
  --env MEMORY_EXTRACTION_ENABLED=true
The six memory types available:
TypePurpose
EpisodicConversation turns and events
SemanticExtracted facts and knowledge
ProceduralLearned rules and procedures
EntityKnowledge graph nodes
Trust-weightedCross-agent trust signals
TemporalTime-bounded facts with expiry

​
Custom Inference Providers

The runtime supports any OpenAI-compatible inference endpoint: 
# Use TrustGate (managed)
PROVIDER_URL=https://trustgate.lucid.foundation

# Use local Ollama
PROVIDER_URL=http://localhost:11434/v1

# Use vLLM
PROVIDER_URL=http://vllm-server:8000/v1

# Use LiteLLM
PROVIDER_URL=http://litellm:4000/v1

​
DePIN Storage

Enable decentralized storage for agent artifacts and memory snapshots: 
DEPIN_UPLOAD_ENABLED=true
DEPIN_PERMANENT_PROVIDER=arweave    # For immutable data
DEPIN_EVOLVING_PROVIDER=lighthouse  # For mutable data
ProviderTierUse Case
ArweavePermanentEpoch bundles, passport metadata, deploy artifacts
LighthouseEvolvingMemory snapshots, MMR checkpoints
MockEitherDevelopment and testing

​
Building a Custom Runtime

If the base runtime does not meet your needs, build your own image and use Path A (BYOI): 
import express from "express";
import { RaijinLabsLucidAi } from "raijin-labs-lucid-ai";

const app = express();
const lucid = new RaijinLabsLucidAi({
  serverURL: process.env.LUCID_API_URL || "https://api.lucid.foundation",
  security: { bearerAuth: process.env.LUCID_API_KEY },
});

// Add your custom logic
app.post('/run', async (req, res) => {
  // Custom preprocessing
  const enrichedPrompt = await myPreprocessor(req.body.prompt);

  // Inference with automatic receipt
  const result = await lucid.run.chatCompletions({
    body: {
      model: process.env.LUCID_MODEL || "openai/gpt-4o",
      messages: [{ role: "user", content: enrichedPrompt }],
    },
  });

  // Custom postprocessing
  const finalResult = await myPostprocessor(result);
  res.json(finalResult);
});

app.listen(3100);
Then deploy: 
lucid launch --image my-custom-runtime:latest --target railway

​
Rollout and Updates

The control plane supports blue-green deployments for safe runtime updates: 
# Deploy new version alongside existing
curl -X POST https://api.lucid.foundation/v1/agents/:passportId/deploy/blue-green \
  -H "Authorization: Bearer lk_..." \
  -d '{"image": "my-runtime:v2.0.0"}'

# Promote after health check passes
curl -X POST https://api.lucid.foundation/v1/agents/:passportId/promote \
  -H "Authorization: Bearer lk_..."

# Or rollback if issues are detected
curl -X POST https://api.lucid.foundation/v1/agents/:passportId/rollback \
  -H "Authorization: Bearer lk_..."