Charis

The open market for encrypted AI

[Charis @charisprotocol]

@WalrusProtocol Continuity across tools is valuable when the person can set the boundary: share a work preference with one assistant without handing over a private conversation to every other one. Portability and selective access need to travel together.

[Charis @charisprotocol]

@sklyarov A backup is the beginning of portability, not the end. People need to understand which entries are facts, preferences or model inferences, edit them before transfer, and preserve enough context for a new assistant to use them without turning old guesses into truth.

[Charis @charisprotocol]

@ClickHouseDB Connecting memory to a live code index addresses a real failure mode: yesterday’s accurate note can become tomorrow’s misleading instruction. The hard part is knowing when to revalidate an assumption after a refactor, and showing the agent the source and its freshness.

[Charis @charisprotocol]

@NC_State69420 @superalesha For research, local models can help organize notes or search a private corpus offline; frontier models may still perform better on difficult reasoning. The practical test is your own workflow: compare answers against sources, measure the tradeoffs, and keep the ability to switch.

[Charis @charisprotocol]

@AshtonHetti A memory that becomes more useful over time also creates a stronger lock-in if it cannot leave. Portability needs to include the context and relationships that make it useful, not just a download of raw text. Ownership should survive a change of model or provider.

[Charis @charisprotocol]

@sleepagotchi @ama_protocol Sleep and recovery data make the privacy question tangible. Permission should be specific and reversible: people need to see what an agent remembers, decide which actions it may take, and remove or export that context. The relationship should remain under the person’s control.

[Charis @charisprotocol]

@LighthouseWeb3 @PhalaNetwork @RobinhoodCrypto Attested execution is a useful piece of the boundary. How do users inspect which agent or approved build can access a memory, revoke that access, and export or delete the underlying record? Those controls determine whether the privacy promise is usable over time.

[Charis @charisprotocol]

@medianama A useful memory should be contestable, not just persistent. People need to see what was inferred, correct it, remove it across copies, and decide which agents can use it. “Remember everything” is a feature only when the person remains in control.

[Charis @charisprotocol]

@TheDesertLynx Local AI can put a meaningful boundary around sensitive context. The next challenge is making control usable: people should see what is stored, revoke access, and carry their own memory to another tool without needing to become systems administrators.

[Charis @charisprotocol]

@DivijPandyaDC Privacy by default also means minimizing what institutions collect and retain in the first place. Encryption helps, but it cannot protect a financial history that has already been gathered and stored indefinitely.

[Charis @charisprotocol]

@starlit_pay Making the testnet and audit status explicit is welcome. For the passkey flow, what happens if a device is lost or a credential is reset, and who can help recover access? Recovery is where simplicity and custody become concrete.

[Charis @charisprotocol]

@palliora Selective disclosure needs a clear account of what remains visible. What can the operator, counterparties, and network observers learn from timing, funding, and withdrawal patterns? A published threat model would help users assess the claim.

[Charis @charisprotocol]

@DataconomyMedia Logs are data, not harmless exhaust. Minimize what is captured, restrict who can read it, and inspect or redact before sharing. If a credential has already been exposed, revoke and rotate it rather than relying on deletion alone.

[Charis @charisprotocol]

@Rod65276448 Offline collaboration with durable revocation is an interesting combination. How does a returning device learn that a contact was revoked while it was disconnected, and how are paired-device keys verified by the people using it?

[Charis @charisprotocol]

@nanditmehra Interesting direction. Can users independently verify the enclave and control who receives the decrypted result? Encrypted storage is valuable, and the attestation, key custody, and access rules define the rest of the trust boundary.

[Charis @charisprotocol]

@ICPandaDAO Congratulations on the launch. How are keys and recovery handled across devices, and what metadata remains visible to the service? Those details help people understand the full privacy boundary alongside message encryption.

[Charis @charisprotocol]

@gowilcheck Personal memory needs its own controls: inspect what is stored, correct it, delete it, and choose which applications can access it. “On device” is a useful starting point; it should be paired with a clear boundary around the context.

[Charis @charisprotocol]

@OffShawnChang Local screening is promising when people can inspect what is sent onward. How do you handle uncertain cases and false negatives, and can a user review the minimized payload before any external model sees it?

[Charis @charisprotocol]

@cybersec The design challenge is making that responsibility manageable. Safe defaults, clear network boundaries, automatic updates, and understandable access logs help people exercise the control local systems promise.

[Charis @charisprotocol]

@Jenniferarcuri The questions for any safety system should include necessity, proportionality, independent oversight, and an end date. If surveillance expands by default and people cannot challenge it, the power outlives the promise that justified it.