The $6 Pico W Is Now a Full Bluetooth Proxy. Three Slots Cost 17KB of Its 264.

ESPHome turned the $6 Pico W into a full Bluetooth proxy with three connection slots already switched on. Here is what they cost in RAM, and how to change it.

Flat illustration of a small green board with three glowing links radiating outward and a segmented memory bar beside it, mostly empty
12 min read
2,478 words

Flash a Raspberry Pi Pico W as a Bluetooth proxy today and Home Assistant does not just listen through it. It can open connections through it — read a thermostat, write to a lock, subscribe to a sensor that only talks when spoken to. Three of those connections at once, switched on before you type anything. That is a new capability on a board Raspberry Pi sells for $6, and setting up a Pico W Bluetooth proxy for Home Assistant now commits about 17KB of static RAM to those three slots whether you use them or not.

The RP2040 at the centre of that board has 264kB of RAM in total. So the default costs roughly one part in fifteen of everything the chip has, and nothing in the ESPHome dashboard will mention it as you click through.

That is worth understanding before you flash, because the number is adjustable and almost nothing written about this change says so.

What ESPHome 2026.8.0 actually gave the Pico W

The release is ESPHome 2026.8.0, published on 19 August 2026, and the distinction it introduced is the one the whole topic turns on.

Until then, a cheap Bluetooth proxy meant a relay. Small BLE devices — a temperature sensor, a plant monitor, a BTHome button — shout their readings into the room every few seconds whether anyone is listening, and a relay simply repeats those shouts to Home Assistant over Wi-Fi. That is enough for a large share of what people actually own, and it needs nothing clever from the board.

A full proxy does something harder. Some devices say nothing useful until you connect and ask: locks, many thermostats, anything with a configuration you can change. Holding that conversation means running a GATT client, and holding it open means reserving memory for it. ESPHome’s 2026.8.0 release notes put the Pico W in that second category for the first time, saying it “supports full active connections (#18132) and now offers 3 connection slots with ESP32 parity (#18247), so Home Assistant can read, write and subscribe to BLE devices through a Pico W proxy.”

ESPHome release notes naming the new rp2_ble_tracker for the Pico W and Pico 2 W, and stating that the Pico W supports full active connections and three connection slots with ESP32 parity
ESPHome's 2026.8.0 release post, published 19 August 2026 and captured 17 Nov 2026: the Pico W gains full active connections in pull request #18132 and three connection slots in #18247.

The same release added the component that makes it possible, rp2_ble_tracker, which ESPHome describes as scanning “passively and actively while WiFi holds an API connection” on the Pico W and Pico 2 W. The proxy sits on top of it and will not build without it. Underneath both, the connection work is handled by a BTstack GATT client, which ESPHome’s documentation names as the mechanism behind active connections on these boards.

One correction worth making in passing, because it is repeated widely. XDA’s write-up of this change, published 15 November 2026, frames it as ESPHome previously being unable to use the Bluetooth chip on these boards at all. That is not quite right: the 2026.3.0 changelog already carried an entry adding a BLE component for the RP2040 and RP2350. The radio was reachable months earlier. What did not exist was a tracker, and above it a proxy, which is a meaningfully smaller claim than the chip being dark.

Two switches are both called active, and they do different things

This is where the topic generates more confusion than anything else about it, and ESPHome’s documentation heads it off in a single clause that almost nobody quotes. Describing the proxy’s active option, the Bluetooth Proxy page notes that it “is separate from active scanning (configured in your platform’s BLE tracker).”

Two settings, same word, different costs, and both on by default.

The first is active under bluetooth_proxy. It governs whether Home Assistant may open GATT connections through the board at all. Turning it on is what reserves the connection slots, so its price is paid in the Pico W’s RAM. Set it to false and you have an advertisement-only relay again — still useful, still fine for BTHome sensors, simply unable to reach a lock.

The second is active under scan_parameters in rp2_ble_tracker, and it has nothing to do with connections. ESPHome’s tracker documentation says scanning is active by default, meaning the board “sends scan requests asking devices for more data” after hearing an advertisement, then merges the reply into what it forwards. That extra question costs nothing in RAM. It costs battery on the devices being asked, and radio time on the Pico W’s CYW43439 chip, which the documentation notes is shared between Wi-Fi and Bluetooth. ESPHome says turning it off “saves power on the advertising devices and radio time on the shared CYW43439.”

Knowing which is which turns a vague complaint into a specific fix. A coin-cell sensor that used to last a year and now does not is a scanning problem, so reach for the tracker. A board that compiles but behaves badly once you add components is a memory problem, so reach for the proxy.

The tracker’s scan defaults are worth seeing once, because they are deliberately conservative rather than aggressive: a 100ms interval, a 30ms window, a five-minute duration, which ESPHome describes as a 30% duty cycle chosen “to leave the radio mostly free for WiFi.” On a board where one antenna serves both, that restraint is the reason your proxy does not also become your Wi-Fi problem.

What three connection slots cost on a 264kB board

Here is the figure, in ESPHome’s own words. On the Bluetooth Proxy page, under connection_slots: “On RP2040/RP2350 the maximum and default are both 3; three slots use about 17KB more static RAM than a single-slot build.”

ESPHome documentation listing the active option as defaulting to true on ESP32 and RP2040/RP2350, and connection_slots with a maximum and default of 3 on RP2040/RP2350 costing about 17KB more static RAM
ESPHome's Bluetooth Proxy documentation, captured 17 Nov 2026 from docs version 2026.10.2: active defaults to true on RP2040/RP2350, and three connection slots cost about 17KB more static RAM than a single-slot build.

Raspberry Pi’s own Pico-series documentation gives the other half of the sum: the Pico W’s RP2040 carries 264kB of on-chip SRAM. Divide one by the other and the default is consuming about 6.4% of the chip’s entire memory before a single sensor, light or button is defined. On the Pico 2 W, whose RP2350 has 520kB, the same 17KB lands nearer 3.3% — the identical feature, half the relative weight, which is a real argument for spending $7 instead of $6 if you intend to put much else on the board.

The comparison with the ESP32 is where this stops being a footnote. The same ESPHome page says that on an ESP32 “each configured slot consumes ~1KB of RAM, with a maximum of 9”. Three slots there cost about 3KB; three slots on a Pico W cost about 17KB. So the Pico W is not merely capped lower at three — its slots are individually far more expensive, by something close to a factor of six. People who have run ESP32 proxies for years and assume slots are nearly free are carrying an intuition that does not transfer to this board.

What makes the default genuinely easy to reconsider is that for a lot of households those slots sit idle. ESPHome is explicit that passively broadcast sensor data “is received separately and is not limited by the number of connection slots” — BTHome, Xiaomi, RuuviTag, Mopeka and the rest of the advertisement-based ecosystem never touch one. If the devices you own only broadcast, three slots buy you nothing and cost 17KB. Setting connection_slots: 1 or dropping to active: false hands that memory straight back, and the documentation names reclaiming static RAM as exactly what lowering the value is for.

If you do have devices that connect, do not over-correct either. ESPHome points out that devices connecting briefly to exchange data and then disconnecting free the slot again, so a handful of occasional devices can share fewer slots than their number. It is the ones that hold a connection permanently — some locks, some thermostats — that each pin a slot for as long as they are paired. Three is a ceiling on simultaneity, not on how many devices you may own, which is a distinction worth getting right before you decide the board is too small. It is the same shape of question as how little hardware local speech-to-text actually needs: the answer is usually more than the spec sheet suggests, provided you know which number is the real constraint.

Set up a Pico W Bluetooth proxy for Home Assistant

You need a Pico W, which Raspberry Pi sells at $6, or a Pico 2 W at $7 if you want the larger memory budget. Retail varies — XDA puts it at $6 or $7 depending on the vendor. You also need a USB cable and a working ESPHome installation.

Flashing is the easy part. XDA’s walkthrough uses ESPHome Web over USB in Chrome or Edge, which avoids installing anything locally, and that is the path to take for a first board. Worth knowing in advance: the Pico W is not among ESPHome’s ready-made Bluetooth proxy projects, so there is no one-click image for it. You will be editing YAML.

Once the device is adopted, the proxy itself is two blocks added to what the dashboard generated for you:

rp2_ble_tracker:

bluetooth_proxy:

That is a complete, working full proxy — three slots, active scanning, everything on. To take the defaults apart, be explicit:

rp2_ble_tracker:
  scan_parameters:
    active: false      # stop asking devices for extra data; saves their battery

bluetooth_proxy:
  connection_slots: 1  # reclaim most of the 17KB; keep one slot for a lock

Or, for a pure relay that reserves no connection memory at all:

bluetooth_proxy:
  active: false

There is one more pattern in ESPHome’s tracker documentation worth copying if the board is doing anything besides proxying. Because scanning runs while Wi-Fi holds an API connection, you can tie it to Home Assistant actually being there:

rp2_ble_tracker:
  scan_parameters:
    continuous: false

api:
  on_client_connected:
    - rp2_ble_tracker.start_scan:
        continuous: true
  on_client_disconnected:
    - if:
        condition:
          not:
            api.connected:
              state_subscription_only: true
        then:
          - rp2_ble_tracker.stop_scan:

The state_subscription_only condition matters more than it looks: without it, a log viewer or a CLI session disconnecting would stop your scan, because every departing client fires that trigger.

The upgrade warning in the docs, and who it really applies to

ESPHome’s Bluetooth Proxy page carries an upgrade notice, and it reads alarmingly: “RP2040/RP2350 proxies were advertisement only before active connection support landed, so existing Pico W configurations that omit this option will gain connection slots on their next update; set active: false to keep an advertisement-only proxy.”

Read quickly, that sounds like a fleet of quiet relays about to start reserving memory. Check it against the release history and the picture narrows considerably.

The advertisement proxy on this platform and its active connections arrived in the same release. ESPHome’s 2026.8.0 notes list the platform-neutral advertisement proxy as #17880 and the Pico W’s active connections as #18132, and that same release first shipped rp2_ble_tracker, which the proxy requires in order to build. The release before it that touched this area, 2026.3.0, added only the lower-level BLE component for the RP2040 and RP2350 — no tracker, no proxy. Which means that on the stable channel there was never a version where you could run an advertisement-only Pico W proxy and then have it change under you. Both halves shipped together.

The warning is not wrong; it is narrow. It applies to people who tracked the beta or dev channel inside the 2026.8.0 cycle, between the pull request that enabled the proxy and the one that enabled connections on it.

For everybody else — which is to say almost everyone reading this — nothing is silently changing. You are making a first-time choice that has already been made for you, and the only thing standing between you and a different choice is knowing the setting exists. That is the same quiet pattern as Home Assistant’s own exposure defaults, where a local model can reach your garage door before you change anything: the default is defensible, documented, and almost never read.

What still does not work, and when to buy an ESP32 instead

A Pico W proxy is a proxy. It is not yet a presence sensor. ESPHome’s tracker page states that BLE sensor platforms such as ble_presence “do not yet bind to this tracker”, with support arriving as those platforms migrate to the shared BLE layer in later releases. If your plan was room presence on a $6 board, that is still ahead of you.

The name also oversells the scope. Despite being called bluetooth_proxy, ESPHome notes that only BLE devices and their Home Assistant integrations are supported — classic Bluetooth is not proxied on any platform.

Two ESP32-only options will bite if you copy a config across. cache_services, which speeds up active connections by caching GATT services in flash, is ESP32-only, and ESPHome is clear that setting it on another platform “is a configuration error” rather than something quietly ignored. The connections list, for configuring slots individually, is ESP32-only as well.

Then there is the ceiling. Three slots is the hard maximum on the Pico W and Pico 2 W where an ESP32 allows up to nine, with ESPHome recommending no more than five for stability. If you genuinely need many simultaneous connections, the cheaper board cannot be talked into it.

And if your actual problem is range rather than cost, the recommendation is unchanged. The Pico W is Wi-Fi only, with no Ethernet and no external antenna connector, and ESPHome’s advice for improving reception is precisely to use a board with Ethernet to keep Bluetooth off the Wi-Fi radio, and ideally an external antenna. Those are ESP32 features. The Pico W’s case is cost and availability — a spare board in a drawer becoming a useful node — not reach.

Whichever board you end up with, one piece of ESPHome’s placement advice applies to all of them and costs nothing: keep the node away from racks, routers and switches, at least three metres, because the interference will undo whatever you gained in configuration. The people most likely to ignore it are the ones already running a rack — the self-hosters who keep their AI on their own hardware are exactly that crowd — because the proxy ends up beside the switch simply where the power strip is.

Sources

Every figure in this article traces to one of these. Dates are when the page was read.

  1. Bluetooth Proxy
    ESPHomeAccessed Nov 17, 2026Official docs
  2. RP2 BLE Tracker
    ESPHomeAccessed Nov 17, 2026Official docs
  3. ESPHome 2026.8.0: Bluetooth on more chips than ever
    ESPHomeAccessed Nov 17, 2026Official docs
  4. ESPHome 2026.8.0 changelog
    ESPHomeAccessed Nov 17, 2026Official docs
  5. ESPHome 2026.3.0 changelog
    ESPHomeAccessed Nov 17, 2026Official docs
  6. Pico-series microcontroller boards
    Raspberry PiAccessed Nov 17, 2026Official docs
  7. Raspberry Pi Pico W: your $6 IoT platform
    Raspberry PiAccessed Nov 17, 2026Vendor pricing
  8. Raspberry Pi Pico 2 W on sale now at $7
    Raspberry PiAccessed Nov 17, 2026Vendor pricing

Written by

Kai Devlin AI Automation Engineer

Full-stack developer who builds and runs AI automation systems in production. Runs local LLMs on personal hardware, builds N8N pipelines that actually ship, and deploys on Cloudflare Pages. Every guide on Pipeline Monk is tested on real consumer hardware — a Ryzen 7 5800HS with 16GB RAM and a GTX 1650. If it works on that, it works.