Status: 25 September 2026
Scope: routing/pathfinding, channel selection, fee control, circular rebalancing, swaps, liquidity estimation, and high-signal research.
Evidence labels: deployed = shipping software; operational tool = runnable against real nodes; simulation/research = evaluated in a model, simulator, or graph snapshot.
- Today's strongest deployed “intelligence” is probabilistic routing plus rules and optimization—not autonomous AI. LND Mission Control, Core Lightning Askrene/XPay, LDK scoring, and Eclair's pathfinder learn from failures or score routes under uncertainty, but remain constrained numerical systems.
- Automatic balancing is four different activities: circular self-payments, submarine swaps, fee shaping, and capital/channel changes. Circular rebalances are cheapest when a good route exists; swaps are more dependable but add provider and on-chain costs; fee changes influence future flow but do not create liquidity; opening channels commits new capital.
- The most complete LND automation suite is LNDg; the most mature composable CLN stack is Askrene/XPay plus Sling/xrebalance/feeadjuster. CLBOSS goes further by combining channel opening, swaps, rebalancing, and fees, but its broad authority requires stronger controls.
- Channel selection is moving beyond centrality. Historical autopilots favor degree or betweenness. Newer GNN/RL work uses capacities, policies, graph bottlenecks, and sequential budget decisions. MPFlow is the clearest production-connected example, but it optimizes modeled max-flow—not demonstrated net profit.
- Fee automation remains mostly heuristic. Rules based on balance, flow, failed HTLCs, peer pricing, and cost recovery are deployed. RL fee agents exist mainly in simulators. A safe production path is bounded fee experiments followed by causal measurement, not immediate RL control.
- The operator objective must be net return, not balanced channels or raw volume: forwarding and lease revenue minus rebalance, swap, chain, infrastructure, and capital costs.
| Implementation | Deployed routing/liquidity mechanism | Operator implication |
|---|---|---|
| LND | Mission Control stores amount-conditioned successes/failures. The apriori and bimodal estimators decay stale evidence and convert it into route probabilities. LND Autopilot still uses graph heuristics such as preferential attachment and betweenness for peer selection. | Strong local payment-attempt evidence, but no exact remote-balance map and no complete routing-node economics controller. |
| Core Lightning | Askrene represents liquidity constraints in layers; XPay plans and retries payments using those layers and learned outcomes. RenePay, the earlier Pickhardt-style payer, was deprecated in v26.06 and is scheduled for removal in v27.03. | The most composable current base for CLN routing experiments, probing layers, and min-cost-flow rebalancing. |
| LDK / rust-lightning | Probabilistic scoring penalizes channels using historical failures/successes and liquidity estimates; route construction and scoring are library components that wallet builders can customize. | Good embedded routing primitives, but the application must supply persistence, telemetry, and operating policy. |
| Eclair | Production pathfinding balances fees, probability, CLTV, and path constraints; supports trampoline routing for clients that delegate parts of route construction. | Strong payer routing; not a routing-node auto-rebalancer or fee-profit controller. |
A 2024 comparative study, An Exposition of Pathfinding Strategies Within Lightning Network Clients, found materially different fee, reliability, hop, and timelock trade-offs among clients. Its broader lesson remains current: there is no single “best route” independent of objective and uncertainty.
- LNDg — operational, broad automation. Uses forwarding/payment history for fee suggestions, bounded Auto-Fees, profitable-route-oriented circular Auto-Rebalancing, peer suggestions, P&L, and optional AR autopilot. It can mutate fees and send rebalance payments, so it needs an admin-level trust boundary.
- Lightning Loop / AutoLoop — deployed swap automation. Loop Out converts outbound Lightning liquidity to on-chain/inbound capacity; Loop In refills outbound capacity. AutoLoop applies rules automatically. More reliable than finding circular routes, but swap, miner, timing, and service costs matter.
- Boltz Client — operational unattended swaps for LND and CLN. Automates swap-based balancing through the Boltz service. Similar economic caveat: it buys a balance change rather than earning one through organic flow.
- Balance of Satoshis — active operator toolkit. Mature LND CLI with circular rebalancing, probing-assisted route work, accounting, fee and peer utilities. Powerful for an operator or scripts, but not itself a complete economics controller.
- regolancer — active circular rebalancer. Automatically chooses source/target channels, calculates amounts and fee ceilings from expected economics, retries routes, and can optionally probe. More automation than BoS, but probing and rebalance payments are active network actions.
- charge-lnd — maintained rule-based fee manager. Applies channel policies from balance, activity, peer, cost-recovery, and other criteria; supports inbound-fee fields. Deterministic and auditable, but rules must be designed and monitored by the operator.
- Lightning Terminal AutoFees — deployed product feature. Automates LND fee changes through Lightning Labs' integrated suite. Public information establishes the feature, but exposes less algorithmic detail than open policy engines such as charge-lnd or LNDg.
- CLBOSS — active broad autopilot. Heuristically opens channels, acquires inbound capacity through Boltz, invokes xrebalance, and sets competitive fees. This is the closest open-source “manage my routing node” package, but it intentionally controls funds and policies.
- Sling — active automatic circular rebalancer. Runs persistent push/pull jobs with target ratios, fee/hop limits, candidate sets, parallelism controls, and learned liquidity information.
- xrebalance — active Askrene-based executor. Uses node-splitting to turn circular self-payments into Askrene min-cost-flow problems, can split across sources/destinations, retries after failures, stores temporary liquidity constraints, enforces fee budgets, and supports dry runs. It deliberately leaves what/when/why to a higher-level controller.
- lightningd/plugins — active curated collection. Includes
rebalance,circular,sling, andfeeadjuster. FeeAdjuster changes fees according to balance/forwarding signals; the rebalancers vary from one-shot operation to background jobs. - LNRadar — active measurement tool, not a rebalancer. Sends deliberately unpayable CLN probes, derives upper/lower liquidity bounds, and writes them into an Askrene layer for XPay. It can improve route beliefs, but consumes remote HTLC/liquidity resources and should remain explicitly permissioned and rate-limited.
- Lightning Pool offers non-custodial channel-lease auctions: it prices inbound capacity rather than rebalancing existing channels.
- Commercial marketplaces such as Amboss Magma also sell channel liquidity. They may solve an inbound-capacity problem, but are not substitutes for route selection, fee optimization, or profitability accounting.
“Latest commit” means the latest commit on the default branch observed on 25 September 2026; it may be a dependency or documentation commit. Not archived does not by itself mean actively maintained.
| Repository | Latest default-branch commit | Archived? | Assessment |
|---|---|---|---|
| lightningnetwork/lnd | 2026-09-25 | No | Active core implementation |
| ElementsProject/lightning | 2026-09-24 | No | Active core implementation |
| lightningdevkit/rust-lightning | 2026-09-23 | No | Active mirror; development is on git.rust-bitcoin.org |
| ACINQ/eclair | 2026-09-23 | No | Active core implementation |
| lightninglabs/loop | 2026-09-24 | No | Active/deployed |
| BoltzExchange/boltz-client | 2026-09-18 | No | Active/deployed client |
| alexbosworth/balanceofsatoshis | 2026-09-23 | No | Active |
| cryptosharks131/lndg | 2026-07-26 | No | Active |
| rkfg/regolancer | 2026-06-06 | No | Active |
| accumulator/charge-lnd | 2026-02-19 | No | Maintenance activity; last recent commits mostly dependencies/docs |
| lightninglabs/lightning-terminal | 2026-09-25 | No | Active/deployed suite |
| ksedgwic/clboss | 2026-09-24 | No | Active; maintainership transferred in 2025 |
| daywalker90/sling | 2026-09-15 | No | Active |
| ksedgwic/xrebalance | 2026-09-08 | No | Active, requires modern CLN |
| lightningd/plugins | 2026-09-21 | No | Active curated plugin collection |
| Lagrang3/lnradar | 2026-04-27 | No | Active operational prototype |
| lightninglabs/pool | 2026-05-22 | No | Maintained liquidity market |
| C-Otto/rebalance-lnd | 2026-04-14 | Yes | Historical LND rebalancer; final activity was dependency maintenance |
| bitromortac/lndmanage | 2024-01-06 | No | Useful historical design; stale default branch |
| giovannizotta/circular | 2024-06-04 | No | Low standalone activity; circular is also distributed through lightningd/plugins |
| LightningCrashers/LNFee | 2023-01-24 | No | Stale research repository, not production fee automation |
| Year | Work | Evidence and practical meaning |
|---|---|---|
| 2020 | Probing Channel Balances | Testnet empirical. Demonstrated rapid balance-bound probing. Historically important, but probes are privacy-sensitive active traffic and observations stale immediately. |
| 2021 | Payments with Uncertain Channel Balances | Simulation/theory. Probability-aware path selection reduced expected attempts in its topology-based simulation; established the value of calibrated uncertainty. |
| 2021 | Optimally Reliable & Cheap Payment Flows | Algorithm + experiments; later influenced deployed CLN work. Treats MPP as probabilistic minimum-cost flow and updates beliefs after outcomes. |
| 2022 | DyFEn and LNFee | Simulation. Multi-channel deep-RL fee-setting environments and baselines; useful research scaffolding, not proof that RL out-earns robust rules in production. |
| 2022 | DRL Rebalancing Policies | Discrete-event simulation. Learns when to use submarine swaps to maximize modeled relay fortune. Valuable formulation; simulator-to-production economics remain unproven. |
| 2023 | Rational Economic Behaviours in the Bitcoin Lightning Network | Economic analysis. Reinforces that forwarding volume and balanced channels do not necessarily cover capital and operating costs. |
| 2024 | Channel Balance Interpolation | Empirical ML with opt-in Amboss data. Roughly 10% better than equal-split prediction on its evaluation; requires future-time and routing-utility validation. |
| 2024/26 | Joint Node Selection and Resource Allocation | Simulation/DRL. Transformer-enhanced policy jointly selects peers and capital. Interesting for channel portfolios, but not demonstrated as a live operator controller. |
| 2025/26 | GNN-Based Autopilot Recommendation | Simulation. GNN/PPO peer recommendation targeting payment success and imbalance; retain the CLoTH/simulation qualifier. |
| 2025 | Pricing for Routing and Flow-Control | Theory + simulation. DEBT control lets channels update congestion-like prices from net flow while senders respond through routing/admission decisions. Promising mechanism, not a drop-in LN daemon feature. |
| 2026 | MPFlow | Graph-snapshot RL + reported production recommendations. Sequentially chooses channel peers to improve modeled max-flow; reports 4,640 opens/267.3 BTC across 30 managed nodes. It does not establish improved net routing profit. |
| 2026 | Stochastic Reset Pathfinding | General path-learning theory/experiments. Models failed paths as episodic resets and studies UCB/Thompson approaches. Lightning is an application example; no LN deployment is shown. |
| 2026 | Sluice | Very recent protocol proposal/snapshot evaluation. Explores pooled channel liquidity with locally enforceable reservations. Potentially important, but it changes the liquidity model and is not available in current LN implementations. |