# OneDroid

> Automated end-to-end testing for software built by AI coding agents, a governed MCP gateway, versioned
> agent memory in your own Postgres, and an open, evidence-gated delivery method. AI agent
> governance, data, and tools that belong to you.

OneDroid (onedroid.ai) builds AI agent infrastructure. It is unrelated to Android and to
the open-source Android projects also called "OneDroid".

Positioning: a coding agent that reports "done" is making a claim, not stating a fact.
OneDroid Argus is automated end-to-end testing for AI-written code, and the part that finds
out. Like any end-to-end test it exercises the running system with nothing mocked (HTTP, the
message broker, the database, web pages in a real browser). Unlike a usual end-to-end suite,
its checks are plain scenarios, not test scripts: a separate tester (a QA engineer or an agent,
never the builder) writes them from the ticket and the acceptance criteria before the build,
and the coding agent's credential cannot read them,
so it cannot pass by guessing. It runs acceptance scenarios the coding agent
never saw against the system it just built, once, under load, and then continuously, and
returns a verdict backed by per-scenario evidence anyone can replay. The scenario set is
hashed and written to a cryptographic ledger before the build starts and revealed in full, with an on-chain anchor, when the tester chooses to reveal after
the verdict, so the holdout is provably prior; keeping it unseen is a separate mechanism
(runner-scoped tokens, folder guards) and OneDroid says so rather than
conflating the two. The model is the swappable part.

Cost: independent testing now costs the same order as the AI coding. In one measured
experiment (https://onedroid.ai/case-study-claude-code-alone-vs-tester), Claude Code built a
feature for $1.97 a run testing its own work; with OneDroid Argus as the independent tester the
coding agent cost $1.01 a run, and the tester one session of about $5.30, which wrote both the build checks and the 13 hidden judge checks
for the feature. The tester is paid once per feature. One ticket, 3 runs per setup. There are no test scripts to write and keep alive: the scenarios are written once per feature, before the build.

Status note: the OneDroid Argus runner and CLI are open source under MIT
(https://github.com/OneDro1d/argus-runner). The agent skills and the MCP server are released.
Anyone can sign up for the evaluation control plane at https://argus-dev.onedroid.ai (a test
environment: no promise that data is kept); an environment of your own is by request.

## Claims and evidence

- [OneDroid's claims, with the evidence behind them](https://onedroid.ai/evidence/onedroid.ai): every checkable claim this site, its docs and its public repository make, where each is made, and what OneDroid has published that backs it. Three verdicts (evidenced, contradicted, no evidence published), references, sha256 fingerprints of the sources as fetched. Checked 2026-10-28, by OneDroid on itself, with the same method as the evidence pages. Markdown twin at https://onedroid.ai/evidence/onedroid.ai.md.

## Docs

- [Quickstart](https://onedroid.ai/quickstart): Three things to try in ten minutes each — connect an agent to OneDroid Synapse with one command and one token, give it memory with OneDroid Engram, run the open-source Dark Factory kits — each ending with a check you run yourself. Free on the hosted shape, whatever the size of your team.
- [Docs — connect an agent](https://docs.onedroid.ai): Quickstart for Claude Code and any HTTP MCP client. Personal access token, the `/agent/mcp` endpoint, the bearer header, token scope and rotation, and the errors people actually hit.
- [How OneDroid Argus works](https://onedroid.ai/argus): Why it is end-to-end testing and what it adds, then the evidence contract in detail — sealing and anchoring the scenario set, running it unseen by the coding agent against a pinned artefact, the reveal, the load profile, continuous verification, the three surfaces, and an explicit list of what OneDroid Argus is not.
- [OneDroid Argus for QA teams](https://onedroid.ai/qa-teams): For QA engineers, SDETs and test-automation leads whose developers build with coding agents. The AI writes the code, QA owns the checks it can't see: plain scenarios written from the ticket before the build, sealed, unreadable by the coding agent, run on the deployed system with nothing mocked, under load and on a schedule. What stays QA's job, and what OneDroid Argus does not claim.
- [Case studies](https://onedroid.ai/case-studies): What happened when OneDroid Argus ran on real systems, with the limits of each test stated.
- [Case study: one modernization, three setups](https://onedroid.ai/case-study-modernization-three-setups): Claude Code ported a decades-old Fortran library to Go alone, with a tester checking the manual, and with a tester checking equivalence with the original. On 109 hidden checks: 104, 81 and 106; the equivalence setup cost 41% less than working alone. What the tester checks decides whether the loop helps or hurts. 3 runs per setup, limits stated.
- [Case study: does the code do what its user guide says?](https://onedroid.ai/case-study-code-vs-user-guide): A decades-old numerical Fortran library against 100 checks written from its own user guide, each quoting the sentence it tests, sealed before the run: 91 held, 9 failed (2 medium, 2 low findings on one rarely used path and in the fine print). Every check shown able to fail with deliberately broken builds; reviewed by a second AI model; the test boundary (what the result does not prove) published.
- [Case study: Claude Code alone vs Claude Code with an independent tester](https://onedroid.ai/case-study-claude-code-alone-vs-tester): One feature ticket, 3 judged runs per setup, with OneDroid Argus as the tester. Both setups passed 13 of 13 hidden checks. The coding agent cost 49% less with the tester; counting the tester's session, the whole cost 41% more. The ticket was too easy to separate quality.
- [Case study: FINOS CALM trades API and MCP demo](https://onedroid.ai/case-study-finos-calm): Seven checks sealed before the run with OneDroid Argus, six generated from the demo's CALM architecture file: 5 passed, 2 failed. The MCP guardrail did not cover lookup by id, and a connection the architecture forbids was open on our cluster, which does not enforce network policies. How anyone can check the recorded result, and what the test did not cover.
- [Case study: OTRS 6.0.30 and Znuny 7.3.7](https://onedroid.ai/case-study-otrs-znuny): One sealed set of 26 checks run three times on each of two releases of a ticketing system with OneDroid Argus compare. Six differences between the releases, what stayed the same, and what the test did not cover.
- [Security & deployment](https://onedroid.ai/security): The three deployment topologies, exactly what OneDroid can and cannot see under each, where the OneDroid Argus runner runs and on whose terms (inside your environment next to the system under test; a scheduled monitor can watch a live system; for an app that moves money, one switch refuses every write the owner has not listed; database reads in read-only transactions, writes only through a scenario's declared cleanup block with the credential you give it; a load ramp runs only against a target its operator listed, with production-looking addresses refused), credential custody, OAuth 2.0 discovery and device flow, and an explicit statement of which compliance attestations are NOT held today.
- [OneDroid Engram](https://onedroid.ai/engram): Portable, persistent, permissioned context for every AI application — hybrid search over a typed object graph, full revision history, in a Postgres you control.
- [Financial services](https://onedroid.ai/financial-services): OneDroid for brokers, asset managers, FCMs and capital-markets firms — one integration point for AI teams, portable context, and the evidence that clears security, supervision and operations review.
- [Dark Factory Toolkit](https://onedroid.ai/dark-factory): Skills, patterns and verification gates for governed, test-first, evidence-gated development loops with AI agents — the written method OneDroid Argus enforces. Free and open source (Apache-2.0) — there is no paid tier. What OneDroid sells around it is help with adoption.
- [Set up the kit](https://onedroid.ai/kit): The kit to build a dark software factory, in four parts ordered by what gives the most for the effort: testing first (OneDroid Argus), then eyes and hands (OneDroid Synapse), memory and context (OneDroid Engram), and the open-source Dark Factory kits as the method that ties them together. For the human, why each part is needed and what it does; for the coding agent, every step with its check, ending with a validation report committed in the user's own private repository. Every link in one place.
- [The method, open source](https://github.com/OneDro1d/dark-factory): Apache-2.0. The stage skills, control loop, enforcement hooks and publish gate, including a self-test that proves each scan class can still fail.

## Blog

- [OneDroid blog](https://onedroid.ai/blog): First-hand notes on testing software built by AI coding agents, governing agent tool access, and agent memory. Every post is served as markdown at `<url>.md`; machine index at https://onedroid.ai/blog/llms.txt, Atom feed at https://onedroid.ai/blog/feed.xml.

## Product

- [OneDroid Argus](https://onedroid.ai/argus): Automated end-to-end testing for software built by AI coding agents.
  Acceptance scenarios are sealed and hash-anchored before the build, executed against a pinned artefact
  the coding agent never tested them on, and revealed after the verdict so any scenario can be replayed by
  a runner the customer controls (open source under MIT: https://github.com/OneDro1d/argus-runner). Three uses on one evidence contract: holdout acceptance,
  under load, and continuous.
- [OneDroid Synapse](https://onedroid.ai/synapse): The governed MCP gateway. Agents connect to one URL;
  every tool call is authenticated, policy-checked, and logged while the hub's activity logging is on (fail-open: a logging failure never blocks a call), to an append-only audit log you own.
  Ships with OneDroid Argus and runs standalone.
- [Home](https://onedroid.ai/): What OneDroid Argus is, the three modes, and the run layer it installs.
- [Enterprise](https://onedroid.ai/enterprise): AI agent governance across an organisation — policy enforcement, audit evidence, a verdict on what your agents built, deployment on your own infrastructure.
- [Pricing](https://onedroid.ai/pricing): The OneDroid Argus runner is free and open source (MIT); the hosted control plane is the priced part. Synapse and Engram are free on the hosted shape, whatever the size of your team, including the event log. Enterprise is an annual licence; the boundary is deployment shape, not seat count.
- [Changelog](https://onedroid.ai/changelog): What actually shipped, with dates.
- [About](https://onedroid.ai/about): Built by Providentia Worldwide, out of regulated enterprise deployments.

## Applications

- [OneDroid Synapse app](https://synapse.onedroid.ai): Sign in to create a hub, connect accounts, and issue an agent token.
- [OneDroid Engram app](https://engram.onedroid.ai): Included with a OneDroid Synapse account.

## Key capabilities

- Holdout acceptance testing against scenarios the coding agent never saw
- Hash-anchored, replayable per-scenario evidence
- Stepped load ramps on the same scenario set, for message brokers and HTTP targets
- Continuous scheduled verification, every 15 minutes at the shortest
- Certificates anyone can check with the open CLI, with no OneDroid account and no contact with OneDroid, through a chain RPC endpoint you choose
- Governed MCP gateway — one URL for all agent tool access
- Per-tool access control, not just per-integration
- Credential custody — agents never hold long-lived upstream tokens
- Append-only event log of every tool invocation, included in the free tier
- Hubs and roles — per-team tool surfaces and membership
- Packs — curated toolsets (tools + skills + knowledge) scoped per hub
- Versioned knowledge store with hybrid vector + keyword + graph search
- Bring your own database for all of the above
- Works with any MCP client and any LLM provider

## Connecting an agent

MCP endpoint for token clients: `https://synapse.onedroid.ai/agent/mcp`
Authentication: `Authorization: Bearer syn_<token>` — a personal access token created in
the OneDroid Synapse dashboard under Manage → API Tokens. Tokens are bound to one user and one hub,
and inherit that user's current role in the hub. OAuth clients use
`https://synapse.onedroid.ai/hub/<slug>/mcp` instead, where the path selects the hub.
Full walkthrough: https://docs.onedroid.ai

## Optional

- [Full site text](https://onedroid.ai/llms-full.txt): Every onedroid.ai page above except the blog (which has its own index at https://onedroid.ai/blog/llms.txt) and the third-party evidence pages, concatenated into one document, generated from the HTML. OneDroid's own claims register is in it once, as part of the home page. Fetch this instead of crawling if you want the whole site in one request. The HTML pages remain authoritative.
- [What is a governed MCP gateway?](https://onedroid.ai/what-is-a-governed-mcp-gateway): The definitional page behind OneDroid Synapse, including where the idea does not earn its keep.
- [Evidence pages](https://onedroid.ai/evidence): For each listed company, every checkable claim its own site makes, where it is made, and what the company has published that backs it: three verdicts, references, sha256 fingerprints, a right of reply. Nothing on them says true or false. Not part of llms-full.txt; each page has a Markdown twin at `<url>.md`.
- [Privacy Policy](https://onedroid.ai/privacy)
- [Terms of Service](https://onedroid.ai/terms)
- [Contact](https://onedroid.ai/contact): michal@onedroid.ai — reaches the team that builds the product. We reply within one business day.
- [Support](https://docs.onedroid.ai/support): how to get help with OneDroid Synapse, OneDroid Engram, OneDroid Argus and the kits — email, GitHub issues, what to include in a report
- [Providentia Worldwide](https://www.providentiaworldwide.com/): The consultancy that builds OneDroid.