AndroidAPS

AndroidAPS (AAPS) is a free, open-source automated insulin delivery (AID) app — an "artificial pancreas system" — that runs on Android smartphones and uses an OpenAPS software algorithm to automate insulin dosing.¹ There is no iOS version; the open-source iPhone equivalent is the separate Loop app. AAPS pairs a phone, a continuous glucose monitor (CGM), and a compatible insulin pump into a closed loop.¹

Algorithm and sophistication. AAPS implements the OpenAPS "oref" algorithm. Its most advanced mode, oref1, adds Super Micro Bolus (SMB) — frequent tiny correction boluses instead of only temporary basal rates — introduced in 2018.² It also offers Autosens, which infers how sensitive or resistant you currently are from glucose deviations and adjusts basal and ISF accordingly, and DynamicISF, which continuously rescales insulin sensitivity from total daily dose and predicted glucose (needing about five days of data).³⁴ This makes its handling of changing sensitivity and partly-announced or unannounced meals among the most capable of any loop.

Supported hardware. AAPS supports an unusually broad pump list for a DIY system: Dana R/RS/-i, Accu-Chek Insight and Combo, Omnipod DASH and Eros, Medtrum Nano/300U, Diaconn G8, EOPatch2, Equil, and certain older Medtronic models (the last two needing an extra relay device).⁵ CGM sources include Dexcom (G5/G6/G7), FreeStyle Libre 2/3, and Eversense, mostly routed through the companion xDrip+ app.⁶

Customizability, community, docs. Because the code is open, every parameter is inspectable and tunable — the defining advantage over commercial AID, which cannot be individualized to the same degree.⁷ AAPS has thousands of users worldwide; a multinational survey of open-source AID users drew 897 respondents across 35 countries.⁸ Documentation is extensive and maintained on a public ReadTheDocs site.¹

Setup complexity and legal status. AAPS cannot simply be downloaded: for legal and safety reasons users must build (compile) the app themselves in Android Studio.¹ After installing, you must complete 11 sequential "objectives" — open loop, low-glucose-suspend, then closed loop and SMB unlock progressively, with SMB at objective 9 — a process that typically takes months.⁹ The project states plainly that it is for informational/educational use, carries no warranty or manufacturer endorsement, and is used "at your own risk."¹ AAPS is not a regulator-approved medical device; in 2019 the U.S. FDA warned against unauthorized diabetes devices, including do-it-yourself automated insulin dosing systems, citing risks of unsafe dosing.¹⁰ That FDA warning is still the operative regulatory position. The professional-guideline position, though, is different — and more supportive. The American Diabetes Association's Standards of Care in Diabetes—2026 (Section 7, published December 2025) recommends, at grade B, that clinicians "support and provide diabetes management advice to people with diabetes who choose to use an open-source AID system" (recommendation 7.27; the same recommendation appeared in the 2025 Standards as 7.29, so this is a settled position rather than a new one).¹¹ The ADA is explicit that open-source AID — its preferred term is OS-AID — is still not regulator-approved; the recommendation is about not abandoning the people who use it. Responsibility that commercial support would carry still falls on the user.

Clinical evidence. Uniquely among DIY loops, the AAPS/OpenAPS algorithm was tested in a randomized controlled trial. The CREATE trial (NEJM, 2022) randomized 97 children and adults to a modified AndroidAPS 2.8 + OpenAPS 0.7.0 system (DANA-i pump, Dexcom G6) versus a sensor-augmented pump; time-in-range rose from 61.2% to 71.2% with AID — an adjusted 14-point benefit (about 3 hours 21 minutes more per day in range), with no severe hypoglycemia or ketoacidosis in either arm.¹² A 24-week continuation phase confirmed durable benefit across pump types and ages over 48 weeks.¹³ Large observational datasets agree: a pediatric survey and a multinational adult/caregiver survey both reported significant HbA1c and time-in-range improvements after starting open-source AID.¹⁴⁸

Real-world follow-up continues to support this, though the newer studies are small. According to PubMed, a one-year single-centre observational study of 27 adults starting AAPS reported time-in-range rising from 67.8% to 84.1% at six months and settling at 79.9% at twelve, with HbA1c falling from 6.6% to 6.1% at six months and 6.0% at twelve, and no severe hypoglycemia or ketoacidosis.¹⁵ In 55 children and adolescents aged 6–19 using AAPS, glycemic outcomes were no different between school days and holidays — a practical reassurance that the system holds up through disrupted routines — and longer night-time sleep was associated with more time in range.¹⁶ Both are observational, single-cohort studies, not trials.

What's coming. Development is continuous and community-driven rather than tied to a release cycle. Recent versions (3.4.x; the latest is 3.4.2.3, June 2026, a bug-fix release) have matured DynamicISF and a Full Closed Loop mode that reduces meal announcements, and broadened pump support (e.g. newer Omnipod, Medtrum and Equil patch pumps).³⁵¹⁷ Because the algorithm is open, advances such as more automatic meal handling typically appear here before commercial systems.

Two things worth knowing if you are comparing AAPS with its cousins. First, Omnipod 5: community work to drive the Omnipod 5 pod from open-source loops is underway, but it is further along in Loop and Trio — the open-source Omnipod 5 project page lists an open beta for those two and states that AndroidAPS support is being worked on with no ETA.¹⁸ A developer interview reported at D-Data/ADA 2026 suggested broader AAPS usability in late 2026 or early 2027, but that is a journalist's report of a conversation, not a project commitment, and no AAPS beta exists today.¹⁹ Second, Auto-ISF: the Auto-ISF settings that iAPS and Trio users talk about are not part of official AndroidAPS. Auto-ISF lives only as a third-party community fork built on top of AAPS; the official app ships DynamicISF instead.²⁰

Glucose values: time-in-range target is 70–180 mg/dL (3.9–10.0 mmol/L).

References

1. AndroidAPS project. Welcome to the AndroidAPS documentation (platform, OpenAPS algorithm, required components, build-it-yourself requirement, and risk/no-warranty disclaimer). AndroidAPS ReadTheDocs (accessed 2026). https://androidaps.readthedocs.io/en/latest/index.html ↩ ↩² ↩³ ↩⁴ ↩⁵

2. AndroidAPS project. OpenAPS features — Super Micro Bolus (SMB), an oref1 feature introduced from 2018. AndroidAPS ReadTheDocs (accessed 2026). https://androidaps.readthedocs.io/en/latest/Usage/Open-APS-features.html ↩

3. AndroidAPS project. DynamicISF (DynISF) — dynamic insulin sensitivity from total daily dose and predicted glucose. AndroidAPS ReadTheDocs (accessed 2026). https://androidaps.readthedocs.io/en/latest/DailyLifeWithAaps/DynamicISF.html ↩ ↩²

4. AndroidAPS project. OpenAPS features — Autosens sensitivity detection. AndroidAPS ReadTheDocs (accessed 2026). https://androidaps.readthedocs.io/en/latest/Usage/Open-APS-features.html ↩

5. AndroidAPS project. Compatible Pumps. AndroidAPS ReadTheDocs (accessed 2026). https://androidaps.readthedocs.io/en/latest/Getting-Started/CompatiblePumps.html ↩ ↩²

6. AndroidAPS project. Compatible CGMs / glucose sources (Dexcom, FreeStyle Libre, Eversense via xDrip+). AndroidAPS ReadTheDocs (accessed 2026). https://androidaps.readthedocs.io/en/latest/Getting-Started/CompatiblesCgms.html ↩

7. Burnside MJ, Lewis DM, Crocket HR, et al. CREATE trial protocol — open-source algorithms allow individualisation of algorithm parameters, unlike commercial AID. J Diabetes Metab Disord (2020). https://pubmed.ncbi.nlm.nih.gov/32837953/ ↩

8. Braune K, Gajewska KA, Thieffry A, et al. Why #WeAreNotWaiting — motivations and self-reported outcomes among users of open-source AID: multinational survey (897 respondents, 35 countries; HbA1c 7.14%→6.24%, TIR 63.0%→80.3%). J Med Internet Res (2021). https://doi.org/10.2196/25409 ↩ ↩²

9. AndroidAPS project. Completing the Objectives — 11 sequential objectives; closed loop and SMB unlock progressively. AndroidAPS ReadTheDocs (accessed 2026). https://androidaps.readthedocs.io/en/latest/SettingUpAaps/CompletingTheObjectives.html ↩

10. U.S. Food and Drug Administration. FDA warns against the use of unauthorized devices for diabetes management (including do-it-yourself automated insulin dosing systems). FDA (2019). https://www.fda.gov/news-events/press-announcements/fda-warns-against-use-unauthorized-devices-diabetes-management ↩

11. American Diabetes Association Professional Practice Committee. 7. Diabetes Technology: Standards of Care in Diabetes—2026 (recommendation 7.27, grade B: "Support and provide diabetes management advice to people with diabetes who choose to use an open-source AID system"). Diabetes Care 2025 Dec 8;49(Suppl 1):S150–S165. https://doi.org/10.2337/dc26-S007 ↩

12. Burnside MJ, Lewis DM, Crocket HR, et al. Open-Source Automated Insulin Delivery in Type 1 Diabetes (CREATE randomized controlled trial). N Engl J Med 2022;387(10):869-881. https://pubmed.ncbi.nlm.nih.gov/36069869/ ↩

13. Burnside MJ, Lewis DM, Crocket HR, et al. Extended Use of an Open-Source Automated Insulin Delivery System: the 24-week continuation phase following CREATE (48 weeks total; treatment effect +12.2% TIR, no DKA or severe hypoglycemia). Diabetes Technol Ther 2023;25(4):250-259. https://pubmed.ncbi.nlm.nih.gov/36763345/ ↩

14. Braune K, O'Donnell S, Cleal B, et al. Real-world use of do-it-yourself artificial pancreas systems in children and adolescents: online survey (209 caregivers, 21 countries; HbA1c 6.91%→6.27%, TIR 64.2%→80.7%). JMIR Mhealth Uhealth (2019). https://doi.org/10.2196/14087 ↩

15. Amuedo S, Antequera-González M, Azriel S. Glycemic outcomes with an open-source automated insulin delivery system in adults with type 1 diabetes: a 1-year real-world observational study (27 adults on AndroidAPS; TIR 67.8%→84.1% at 6 months, 79.9% at 12 months; HbA1c 6.6%→6.1%→6.0%; no severe hypoglycemia or DKA). According to PubMed. Diabetes Technol Ther 2026;28(7):739-745. https://pubmed.ncbi.nlm.nih.gov/41566695/ ↩

16. Li X, Ye X, Tang J, et al. Glycaemic control during school days and holidays in children and adolescents with type 1 diabetes using open-source Android artificial pancreas systems (55 participants aged 6–19; no significant difference between school days and holidays). According to PubMed. Diabetes Obes Metab (2026). https://pubmed.ncbi.nlm.nih.gov/42348355/ ↩

17. AndroidAPS project. Releases — latest 3.4.2.3, published 4 June 2026 (bug fixes; there is no 3.5). GitHub — nightscout/AndroidAPS (accessed 2026-07-14). https://github.com/nightscout/AndroidAPS/releases ↩

18. Nightscout / open-source AID community. Omnipod 5 — open-source AID (open beta listed for Loop and Trio; AndroidAPS support in development, no ETA). nightscout.github.io (accessed 2026-07-02). https://nightscout.github.io/omnipod-five/ ↩

19. Diabetech. Omnipod 5 and Tandem Mobi support headed to Loop, Trio and AAPS — developer interview; AAPS at an earlier stage, broader usability suggested for late 2026 / early 2027. Diabetech (2026-06-09). https://www.diabetech.info/p/omnipod-5-and-tandem-mobi-support-headed-to-loop-trio-and-aaps ↩

20. ga-zelle. autoISF — an AndroidAPS extension for specific glucose behaviours; a third-party fork built on AAPS v3.4.0, not part of the official app. GitHub (accessed 2026-07-14). https://github.com/ga-zelle/autoISF ↩