Current status
This page is the public source of truth for what exists. It is updated when the software changes, and every entry uses one of the four status labels below. If another page seems to say more, this page wins. It was last checked against the product in October 2026.
The work now is to make every public statement match the software, keep keys protected by default, and keep the full test suite passing before the next capability is built. See the Roadmap.
Status labels
| Status label | Meaning |
|---|---|
| Working foundation | Implemented and covered by behavior tests for the stated scope. |
| V1-testnet target | Required for V1-testnet and not yet qualified. |
| Research direction | Being designed or tested as a hypothesis, with no release commitment. |
| External validation required | Needs independent technical, legal, operational, or security review. |
Other pages use these labels with the same meaning. Each label is shown in the same style everywhere: an info note for a working foundation, a caution note for a V1-testnet target, a plain note for a research direction, and a warning for external validation.
What works today
| Area | What exists | Status |
|---|---|---|
| Desktop app | Local-first desktop application with a local API and command line. First run happens in the chat, with guided model and wallet setup. Internal pre-release installers for Windows, macOS, and Linux are built for internal testing only and are unsigned; consumer installers wait until the private beta (M3). | Working foundation |
| Local inference | Manages a local llama.cpp-compatible engine. The owner supplies a model or explicitly approves a download; no model is bundled. NVIDIA CUDA acceleration works; other accelerators are not yet used. Answers are written by the local language model. The intent routers that pick a local action when no language model routes a message are rule-based, not language models. | Working foundation |
| Governed knowledge | Governed source registry, trust ceilings, confirmed folder ingestion, encrypted vault and memory, and answers that show their evidence. | Working foundation |
| Wallet and devices | One owner wallet with a separate key for each device. Wallet keys are sealed by default: OS secure storage on desktop, a passphrase on the command line. Custody is not hardware-backed. | Working foundation |
| Device revocation | A revoked device can no longer sign API calls, proofs, or local validator votes. | Working foundation |
| Backup and recovery | A four-step backup flow, offered during first-run setup and in Settings, creates an encrypted recovery kit, shows its one-time secret, and checks the saved kit and secret before it completes. The kit restores the wallet into an empty workspace on the same machine. Recovery on a second device is not yet verified. | Working foundation |
| Activity timeline | A local, hash-linked record of activity whose newest entry is named by a head signed with the wallet key. Edited, removed, or reordered entries, a cut-off tail, and a missing or forged head are detected. Witnessed checkpoints against a rollback of the whole timeline are not yet built. | Working foundation |
| Right to Exit | Owner-triggered export into an encrypted continuity pack and a separate readable pack, with integrity checks; export never deletes anything. Chats, governed document text, and the activity timeline were added to the export in October 2026, and their end-to-end tests are still being written. Forgotten content is left out. Coverage of every data collection and import on a separate machine are not yet verified. | Working foundation |
| Local protection | Encrypted storage for workspace secrets, vault, memory, and queues; one-time launch codes, content security policy, and host checks on the local API; terminal access off by default. Sealing of every chat-derived copy is being completed in M0. | Working foundation |
| Governed sharing | Grants between owners carry a commitment to the shared location, and the grantee and expiry are enforced. Purpose is recorded but not yet enforced, and task inputs are still sent to the assigned peer. | Working foundation |
| Networking | Signed messages between agents, tested between processes on one machine. Multi-machine networking, encrypted transport, and NAT traversal are not yet built. | V1-testnet target |
| Proofs and finality | Proof of Data and Proof of Computation records, re-execution by independent wallets, and prepare-and-commit finality, all running locally. There is no multi-validator network. | V1-testnet target |
| ASHA accounting | Simulated accounting only. No amount in the software has monetary value, and no product action moves value. | Research direction |
| Mobile | Desktop-side pairing and a device-signed API, tested against a simulated phone. The mobile app is in development; there is no signed mobile build. | V1-testnet target |
| GENOME | A local, privacy-preserving graph projection. There is no network-wide GENOME. | Research direction |
Claims audit
In September 2026 every capability claim in the product README, feature list, and interface was checked against the code and its behavior tests. Claims that overstated the software were corrected, and a remaining interface statement is being fixed in code rather than in wording. In October 2026 the product's intent routers were renamed in every user-visible place as rule-based routers, not language models. This page follows the corrected record.
Not claimed
SENEX does not currently claim:
- a public, permissionless, or multi-validator network;
- live ASHA, or any value, transfer, redemption, or reward in the software;
- differential privacy, anonymity, or any other formal privacy guarantee;
- hardware-backed or non-exportable key custody;
- detection of undeclared common control (Sybil identities);
- publicly distributed or signed installers;
- independent security, cryptographic, or privacy review; or
- artificial general or super intelligence.
V1-testnet qualification
V1-testnet is qualified through fourteen gates, G0 to G13, covering product truth and licensing, durable state, consensus, transport, runtime, the ASHA lifecycle, evidence, custody, privacy, governance, product journeys, cross-platform qualification, security and legal review, and the public testnet itself. The gates are in progress; none has exited yet. See Network transition and upgrade policy for how V1-testnet relates to V1.