惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

博客园_首页
IT之家
IT之家
博客园 - Franky
Stack Overflow Blog
Stack Overflow Blog
宝玉的分享
宝玉的分享
Recent Announcements
Recent Announcements
Engineering at Meta
Engineering at Meta
S
SegmentFault 最新的问题
V
Visual Studio Blog
freeCodeCamp Programming Tutorials: Python, JavaScript, Git & More
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻
Last Week in AI
Last Week in AI
H
Help Net Security
V
V2EX
H
Hackread – Cybersecurity News, Data Breaches, AI and More
量子位
博客园 - 叶小钗
J
Java Code Geeks
博客园 - 【当耐特】
月光博客
月光博客
爱范儿
爱范儿
人人都是产品经理
人人都是产品经理
酷 壳 – CoolShell
酷 壳 – CoolShell
小众软件
小众软件

Hacker News - Newest: "AI"

AI can't read an investor deck AI as an attorney? Student uses ChatGPT, Gemini to sue UW over alleged racial discrimination Hacking MCP Servers in AI Systems – The Rug Pull: Tool Changes After Approval GitHub - MeepCastana/KubeezCut: Free Web based video editor Can AI judge journalism? A Thiel-backed startup says yes, even if it risks chilling whistleblowers Coming soon: 10 Things That Matter in AI Right Now DARPA built an AI to fact-check enemy weapons claims What explains heterogeneity in AI adoption? When AI Meets Muscle: Context-Aware Electrical Stimulation Promises a New Way to Guide Human Movements - Department of Computer Science AI Changed How We Build. It Did Not Change What Matters. Linux rules on using AI-generated code - Copilot is OK, but humans must take 'full responsibility for the… Meta spins up AI version of Mark Zuckerberg to engage with employees Code Mode: Let Your AI Write Programs, Not Just Call Tools | TanStack Blog GitHub - Delavalom/graft: Go framework for building AI agents. Type-safe tools, multi-provider (OpenAI, Anthropic, Gemini, Bedrock), zero vendor SDKs. India's TCS tops estimates, says new AI models did not dent services demand Gen Z's fading AI hype Strong feeling: we are in a folded AI reality GitHub - machinarii/total-recall-catalog: A reference catalog of latest knowledge retrieval, memory & RAG systems GitHub - mensfeld/code-on-incus: Give each AI agent its own isolated machine with root, Docker, and systemd. Active defense detects and stops threats automatically.. Quantization, LoRA, and the 8% Problem: Benchmarking Local LLMs for Production AI Iran war: We spoke to the man making Lego-style AI videos that experts say are powerful propaganda Powell, Bessent discussed Anthropic's Mythos AI cyber threat with major U.S. banks GitHub - immartian/bellamem: Persistent belief-graph memory for AI agents. Retrieves decisive context by importance — not recency, not RAG, not /compact. recursive-mode: The Repo-Native Operating System for AI Engineering After the attack on Sam Altman's home, will AI CEO's go on the offensive? The biggest advance in AI since the LLM Opus 4.6 vs GPT 5.4 One Prompt Unity World Generation Test “AI polls” are fake polls Client Challenge Can AI be a 'child of God'? Inside Anthropic's meeting with Christian leaders
GitHub - Jake1848/conduit: Conduit
conduitbtc · 2026-06-22 · via Hacker News - Newest: "AI"

Self-hosted Bitcoin Lightning payment infrastructure for autonomous AI agents.

Conduit is software tooling you run on your own infrastructure, in front of your own LND node, with your own keys. It gives any AI agent a virtual Lightning wallet, a spending policy, and an API to send, receive, and account for Bitcoin payments — programmatically, with hard guardrails the agent cannot override.

There is no Conduit SaaS: you host it, Conduit never holds your funds and never phones home. At the operator level it's self-hosted — your node, your keys, your rules. The agents you create are virtual sub-balances in a ledger that you, the operator, control: they hold a scoped API key, not a signing key, and you can credit, debit, or sweep them at any time.

Status: v0.8.4 — running live on testnet and mainnet (testnet/regtest plus a live mainnet node; the first real mainnet payment has settled end-to-end — still early and small, single-operator, not production-at-scale). The operator treasury (added in 0.8.3) lets you see accrued platform-fee revenue and withdraw accrued BTC on-chain, gated by a solvency guard (node assets can never drop below what you owe agents). 0.8.4 adds a full audit/red-team pass, the complete operator console, and an agent-pays-over-Lightning demo (see DEMO.md).


How it works

You deploy Conduit (a 5-minute Docker bring-up) against your LND node. Conduit sits in front of LND as a policy + accounting layer:

  • You hold the LND seed and macaroon. Conduit only ever talks to LND over the macaroon you mount into it.
  • You are the operator. The bootstrap API key is your master key to your own system — it mints the scoped keys you hand to your agents.
  • You credit and debit the virtual sub-balances of the agents running on your node — Conduit is custodial for the agents by construction: the agent balances are operator-controlled IOUs in Conduit's ledger, and the underlying sats stay in your channels under your keys.
  • You set a per-transaction platform fee (in sats) that Conduit adds on top of each payment and keeps on settlement — that fee is your revenue, configured by you, never a Conduit cut.

Revenue model

Conduit's built-in monetization is a small, usage-based platform fee in satoshis that the operator who deploys Conduit configures. It is charged on top of each payment, kept when the payment settles, and refunded in full if the payment fails. Sats only — no fiat, no cards, no subscription.

env var default meaning
PLATFORM_FEE_PERCENT 0.5 platform fee, as a percent of the payment amount (0.5 = 0.5%)
PLATFORM_FEE_MIN_SATS 1 floor for the per-transaction platform fee
PLATFORM_FEE_MAX_SATS 1000 ceiling for the per-transaction platform fee

Collected fees are reported at GET /v1/fees (admin) and surfaced in GET /v1/metrics. Set PLATFORM_FEE_PERCENT=0 to run Conduit with no fee.

Repository layout

.
├── website/         Landing page (deployed to conduit.energy)
├── core/            Conduit Core API — FastAPI server you run in front of your LND
├── sdk-python/      Python SDK (`conduit-btc` on PyPI; import `conduit`)
├── sdk-js/          TypeScript SDK (`@conduit-btc/sdk` on npm)
├── mcp-server/      MCP server (`conduit-btc-mcp` on PyPI; `conduit-mcp` command)
├── infra/           Bitcoin Core / LND configs, systemd units, install scripts
├── docs/            MkDocs documentation site
├── docker-compose.yml
└── Conduit_Whitepaper_v1.pdf

The five components

# Component Status Path
1 Bitcoin Core + LND nodes install scripts — you run them on your own host infra/
2 Conduit Core API FastAPI app, runs with mock-LND for local dev core/
3 Python SDK matches the website code panel sdk-python/
4 TypeScript SDK mirrors the Python interface sdk-js/
5 MCP server exposes Conduit as MCP tools mcp-server/

Quickstart (local, mock LND)

# 1. Bring up the API in mock mode (LND_MOCK=true, SQLite, bootstrap key preset)
docker compose -f docker-compose.dev.yml up --build

# 2. Hit the API
curl http://localhost:8000/v1/health

# 3. Use the Python SDK
cd sdk-python && pip install -e .
export CONDUIT_API_KEY=ck_test_dev_root
python examples/quickstart.py

The dev container ships with LND_MOCK=true, an auto-created admin-scoped bootstrap key ck_test_dev_root (your master key in dev), and SQLite at /app/data/conduit.db. No real Bitcoin moves until you point Conduit at a real LND node of your own.

Running against a real node

Conduit is self-hostable today on testnet, regtest, and mainnet — all three run live. Mainnet is now real: a neutrino LND node with a real channel is up and the first real-money payment has settled end-to-end. It is still early and small, so treat a mainnet bring-up as new territory and test on testnet first. The order of operations against your own Lightning infrastructure:

  1. Install Bitcoin Core (pruned) on your host — infra/scripts/install_bitcoind.sh
  2. Install LND (testnet today; mainnet supported) — infra/scripts/install_lnd.sh
  3. Configure UFW — infra/scripts/setup_firewall.sh
  4. Wait for chain sync (1–3 days)
  5. Open channels — infra/scripts/setup_channels.sh
  6. Point the Core API at your LND macaroon and TLS cert (LND_MOCK=false)
  7. Set your PLATFORM_FEE_* values to whatever revenue you want to charge
  8. Deploy with docker compose -f docker-compose.yml -f docker-compose.prod.yml up -d

Mainnet (live)

Mainnet is live but early and small. The deployment today is a single neutrino LND node with one ~20k-sat channel. The first real-money payment has settled end-to-end: 2000 sats sent via Lightning keysend, a 10-sat platform fee (0.5%) collected, 1-sat routing fee, settled in ~15s — with the full lifecycle verified (debit → route → settle → platform-fee capture → refund-on-failure → exact ledger reconciliation). This is a first real-money validation, not production-at-scale; do not assume throughput or battle-testing. The live mainnet API is at https://api-mainnet.conduit.energy.

Full runbook: infra/README.md.

Self-hosted trust model

Conduit is self-hosted by construction — it is software you operate, not a service that holds your money. Be clear about who custodies what:

  • At the operator level you are self-hosted. There is no Conduit SaaS; Conduit never holds your funds and never phones home. The LND seed phrase is yours and is never stored on the VPS by Conduit; the sats stay in channels under your keys.
  • At the agent level Conduit is custodial. Agent balances are virtual IOUs in Conduit's ledger that you, the operator, credit, debit, and can sweep. Agents hold a scoped API key, not a signing key — they never touch a Bitcoin private key.
  • LND gRPC / REST is never exposed publicly — only your Conduit API is
  • The Conduit policy engine evaluates every payment before it reaches your LND and fails closed on any error
  • API keys are bcrypt-hashed at rest and shown to you, the operator, exactly once
  • The bootstrap API key is your master key — guard it like the macaroon
  • All transit is HTTPS; webhook deliveries are HMAC-signed

Conduit is the policy + accounting layer in front of a node you control. If you turn Conduit off, your sats are still in your channels.

See infra/README.md → "Security checklist".

Authorization model (single-operator today)

Authorization in Conduit is scope-based, not per-agent. An API key carries one of three scopes — read < write < admin — and that scope is the only boundary:

  • A read key can read the entire fleet — every agent, balance, and transaction.
  • A write key can act on any agent — send a payment from, or create an invoice for, any agent in the ledger.
  • An admin key can create/delete agents, move balances, manage policies and webhooks, and mint keys across the whole instance.

There is no per-agent boundary: a key is not tied to a specific agent, and no route filters by which key created or "owns" an agent. Agents are an accounting and policy unit, not a security boundary between mutually distrusting parties.

Concretely: Conduit today is a single-operator tool. Hand scoped keys to agents you run, not to third parties you don't trust with each other. If you need hard isolation between tenants, run a separate Conduit instance (and node) per tenant.

Roadmap: multi-tenant, per-agent scoping — where a key is bound to one agent (or set of agents) and cannot read or act on the rest of the fleet — is a planned enhancement. As a first step toward it, create_agent now records the minting key on Agent.api_key_id for provenance. This is provenance only: it does not yet change or enforce any authorization.

License

MIT.