Franklin

Every tool call is itemised. Every token is priced. The wallet is the source of truth — when it hits zero, Franklin stops. No overdraft, no surprise bill, no rate-limit wall at 3am.

Why Franklin

💳  Pay per action

No subscriptions. No "Pro" tier. You fund a wallet once and Franklin spends atomically per API call via HTTP 402. Cheap models cost fractions of a cent. Frontier models cost what they cost. When the wallet is empty, Franklin stops.

🔐  Wallet is identity

No email. No phone. No KYC. Your Base or Solana address is your account. Portable across machines — franklin setup imports an existing wallet in one command. Your sessions, your config, your money.

🧠  55+ models, one interface

Claude Sonnet/Opus 4.6, GPT‑5.4, Gemini 2.5 Pro, Grok 4, DeepSeek V3, GLM‑5.1, Kimi, Minimax, plus NVIDIA's free tier (Nemotron, Qwen3 Coder). Switch mid‑session with /model. Automatic fallback if one provider is down.

The comparison

| | Claude Code | Aider | Cursor | Franklin | | --------------------------------- | -------------- | ------------- | -------------- | --------------------- | | Writes and edits code | ✅ | ✅ | ✅ | ✅ | | Multi‑model support | ❌ Claude only | ✅ BYOK | ⚠️ limited | ✅ 55+ via 1 wallet | | Pricing model | Subscription | BYOK | Subscription | Pay per action | | Identity | Account | API keys | Account | Wallet | | Spend budget cap enforced on‑chain | ❌ | ❌ | ❌ | ✅ | | Pay any API (images, search…) | ❌ | ❌ | ❌ | ✅ via x402 | | Plugin SDK for custom workflows | ❌ | ⚠️ | ❌ | ✅ | | Persistent sessions + search | ⚠️ | ⚠️ | ⚠️ | ✅ | | Start free, no signup | ❌ | ⚠️ BYOK | ❌ | ✅ |

Franklin is the first agent in the Autonomous Economic Agent category — an agent that takes a goal, decides what to spend on, and executes within a hard budget cap enforced by the wallet.

Features

🧠 55+ models via one wallet Anthropic, OpenAI, Google, xAI, DeepSeek, GLM, Kimi, Minimax, NVIDIA free tier. One URL, one wallet, automatic fallback.

💳 x402 micropayments HTTP 402 native. Every tool call is a tiny signed transaction against your USDC balance. No escrow, no refund API, no subscription.

🚦 Smart tier routing Mark steps as free / cheap / premium — Franklin picks the best model per tier, per task. Configurable defaults in franklin config.

🔌 Plugin SDK Core is workflow‑agnostic. Ship a new vertical (marketing, trading, research) without touching the agent loop. See docs/plugin-sdk.md.

💾 Persistent sessions Every turn is streamed to disk with full metadata. Resume any session by ID. Survives crashes, reboots, context compaction.

🔍 Full‑text session search franklin search "payment loop" — tokenised search across every past session. No SQLite, no indexing daemon, just fast.

📊 Cost insights franklin insights — daily spend sparklines, per‑model breakdown, projections. Never wonder where the USDC went.

⚡ Anthropic prompt caching Multi‑turn Sonnet/Opus sessions use ephemeral cache breakpoints (system_and_3 strategy). Large input savings on long conversations.

🛠 12 built‑in tools Read · Write · Edit · Bash · Glob · Grep · WebFetch · WebSearch · Task · ImageGen · AskUser · SubAgent.

🔗 MCP auto‑discovery Drop‑in Model Context Protocol servers from ~/.blockrun/mcp.json. Ships with awareness of blockrun-mcp (markets, X, prediction markets) and unbrowse (any site → API).

🧭 Plan / Execute modes /plan to design read‑only, /execute to commit. No accidental writes while exploring.

🪄 Slash ergonomics /commit, /push, /pr, /review, /ultrathink, /compact, /model, /cost, /wallet, and 20+ more.

Plugin SDK

Franklin is plugin‑first. The core agent doesn't know what a "marketing campaign" or "trading signal" is — it just runs workflows. Adding a new vertical is a single TypeScript file.

import type { Plugin, Workflow } from '@blockrun/franklin/plugin-sdk';

const researchWorkflow: Workflow = {
  id: 'research',
  name: 'Competitor Research',
  description: 'Find and summarise 10 competitors in a given space',
  steps: [
    {
      name: 'search',
      modelTier: 'none',          // pure API call, no LLM
      execute: async (ctx) => {
        const results = await ctx.exa.search(ctx.input.topic, { limit: 10 });
        return { output: `Found ${results.length}`, data: { results } };
      },
    },
    {
      name: 'summarise',
      modelTier: 'cheap',         // bulk work — use GLM or DeepSeek
      execute: async (ctx) => {
        const summaries = await ctx.llm.map(ctx.data.results, (r) =>
          `Summarise in 3 bullets: ${r.text}`);
        return { output: 'Summaries written', data: { summaries } };
      },
    },
    {
      name: 'synthesise',
      modelTier: 'premium',       // final output — use Claude Opus
      execute: async (ctx) => {
        const report = await ctx.llm.complete({
          system: 'You are a strategy analyst.',
          user: `Synthesise these into a 1‑page report: ${JSON.stringify(ctx.data