Cortensor

🛠️ DevLog – Offchain Storage Configurability: Next Step After Privacy Basics Now that the basic privacy v1.0 building blocks are in place, one of the next follow-up items is offchain storage configurability. 🔹 What we’re assessing now - Today, our offchain/S3 configuration is still spread across a few places: - router path, where user/app payloads get uploaded to offchain storage - miner path, where miners resolve URNs from chain/offchain and read payloads back - That’s workable for current testing, but too rigid for the longer-term privacy/data-ownership direction. 🔹 What we want to do - First, assess all the places that need to change. - Second, centralize these offchain storage settings into one place. - Third, make them env/config driven so operators can point to their own S3/object storage. - If nothing is set, it should fall back to the current default S3/offchain path we use today. 🔹 Why this matters - Privacy is not only about encryption keys - it also ties into where the encrypted data lives. - Longer term, this lets: - router/node operators choose their own storage backend - session/app owners keep stronger control over where their data is persisted - privacy + data ownership fit together more naturally This will be one of the next action items in the coming weeks while we continue testing privacy v1.0 flows next week. #Cortensor #DevLog #Privacy #OffchainStorage #EncryptedInference #DePIN

🛠️ DevLog – Tool-Calling Gaps (Open Models) & Offchain Payload UI Been poking at tool calls end-to-end through the PyClaw path, and a couple of things surfaced that we’ll fold into Phase 3 work. 🔹 Open models & tool-calling behavior - Current OSS models (via Ollama) feel intentionally weak on tool calling unless you fine-tune or add heavy prompt scaffolding. - In PyClaw experiments, tools “work” but reliability is not close to what we want for agent workflows. - Short-term path: - Experiment with Ollama system/role prompts that explicitly teach our tool schemas + response shapes. - Explore lightweight fine-tuning / adapters for COR-specific agent flows (router tools, browser tools, etc.) so open models learn to call tools predictably. 🔹 Offchain payloads: infra ready, dashboard lagging - The router + SDK already support payload v0/v1/v2, but the dashboard UI only exposes v0 (fully onchain) today. - Next UI work: - Add v1 mode – input offchain, result onchain. - Add v2 mode – input and output offchain (ideal for private / encrypted settings). - Goal is to bring the dashboard up to parity with the infra so agent apps can actually use these paths without manual wiring. 🔹 Takeaways / next actions - $COR infra needs agentic features at the prompt/completion layer so open models behave more like “tool-native” agents, not generic chatbots. - Immediate TODOs: - Add dashboard support for offchain payload v2 (and v1) on sessions. - Iterate on Ollama system/role prompt templates to harden tool-calling for open models used by PyClaw and router-node endpoints. These sit alongside the ongoing privacy work and PyClaw design as additional Phase 3–oriented experiments. #Cortensor #DevLog #PyClaw #ToolCalling #Offchain #AgenticAI #DePIN

🛠️ DevLog – Offchain Data Storage Will Become Configurable One more item surfaced while iterating on the privacy flow: we need to make offchain payload storage more configurable. 🔹 Current limitation - Right now, offchain payloads are stored in our own default S3 path / bucket setup. - That works for testing, but it’s too rigid for the longer-term direction. 🔹 What we want to change - Move offchain storage settings into config / env options so each operator can point to their own S3 bucket or object-storage setup. - This makes the storage layer more portable and less tied to one default infra path. 🔹 Why it matters - Longer term, the goal is: - operators control their own offchain storage - users/apps keep stronger control over their data - and with encryption, storage ownership + data privacy fit together more cleanly This is separate from the current privacy v0.5 testing, but it’s clearly part of the same broader direction around data ownership and private inference. #Cortensor #DevLog #OffchainStorage #Privacy #EncryptedInference #DePIN

🗓️ Weekly Focus – Testnet Phase #3 Start, Stats Reset & Node Restructuring Testnet Phase #3 has started. This week focuses on starting from a clean slate, gathering fresh node data, and preparing the infrastructure needed for the next round of testing. 🔹 Phase #3 Start + Node Stats Reset - Phase #3 begins with a one-time reset of phase stats/reputation so routing and reward data reflect this phase only. - This gives us a clean baseline for measuring node behavior and performance in the new phase. 🔹 Fresh Node Data → Node Restructuring - With the updated node stats coming in after the reset, we’ll begin the next round of node restructuring across environments and roles. - Goal: better alignment between node capacity, roles, and Phase #3 testing priorities. 🔹 Dedicated Node Sessions for Agent Iteration - A subset of nodes will move into dedicated sessions to support: - the 4 main router endpoints (Corgent + Bardiel) - private / encrypted inference iteration 🔹 /delegate & /validate Session Configs - Prepare more stable 1 / 3 / 5 node session configurations for /delegate and /validate. - This enables redundancy and consensus-style testing on predictable node sets instead of only the general pool. 🔹 Private Inference v0.5 – Endpoint Upgrades & Prep - Upgrade all four router endpoints (Corgent + Bardiel) to the latest build so encryption flows can be tested there. - May begin the final E2E integration/testing pass for Private Inference v0.5 this week. - Focus is on dedicated private sessions, ACL-backed access, payload key issuance, and decrypt → infer → re-encrypt flow. Phase #3 starts with infrastructure alignment first - reset stats, restructure nodes, upgrade router endpoints, and prepare the environment for deeper testing on agent primitives and private inference. #Cortensor #Testnet #Phase3 #AIInfra #DePIN #Corgent #Bardiel #Delegate #Validate #PrivateAI #L3

🛠️ DevLog – This & Next Week: Real E2E Privacy Testing on Testnet0/Testnet1a We’ll be spending this week and next week doing more real end-to-end testing of the privacy flows on testnet0 and testnet1a. 🔹 What we’ll be doing - Run more full privacy E2E cases across both environments: - dedicated private sessions - ephemeral private sessions - Keep adjusting router / endpoint setup so the encryption-key flows are properly exposed and testable. - Tighten dashboard UX and fix rough edges as they surface during real usage. 🔹 What we expect during this phase - Some gaps will likely show up around: - router ↔ miner ↔ dashboard integration - privacy UX / key fetch / cache flows - regression around existing non-private paths - We’ll use this time to fix those gaps and refine the flow instead of pushing wider too early. 🔹 In parallel - We’ll also keep a light eye on the next storage-related item: - making offchain payload storage more configurable over time, so privacy can align better with operator-controlled data paths. So the focus for now is simple: test the privacy flows for real on testnet0/testnet1a, fix what breaks, and smooth out the UX/integration path. #Cortensor #DevLog #Privacy #EncryptedInference #Testnet0 #Testnet1a #DePIN

🧩 New Doc – /delegate & /validate v3 Spec (Explicit Redundancy & Consensus) We’ve published a new router-node spec for the next version of Cortensor’s two core agentic rails, /delegate and /validate, focused on explicit redundancy and consensus across miners: 🔹 What v3 adds - Makes redundancy a first-class knob – tasks can target 1, 3, 5+ miners instead of “single session only,” with clear policy fields for redundancy and consensus. - Defines how the router picks independent miners, runs them in parallel, and surfaces consensus metadata (agreement, dissent, retries) back to the caller. - Extends both /delegate (execute under policy) and /validate (verify before acting/paying) so agents can ask for “fast,” “balanced,” or “oracle-grade” profiles instead of hand-rolling trust logic. 🔹 Why it matters - Hardens Cortensor’s agentic surface: agents talk to a clean v3 contract, while the router + miners handle redundancy, PoI/PoUW checks, and aggregation under the hood. - Sets the implementation target for Phase #3, where we’ll move from today’s v2 single-session flows toward explicit redundant inference and richer verdicts for agent workflows. 🔹 Read the spec /delegate & /validate v3 – Explicit Redundancy & Consensus: docs.cortensor.network/community-and-… #Cortensor #DevLog #Delegate #Validate #AgenticAI #DePIN

🛠️ DevLog – Latest Privacy Dashboard Changes Live on Testnet0 We’ve deployed the latest dashboard privacy changes to testnet0 so it now supports both: 🔹 Dedicated private sessions - Current v0.5 dedicated-node privacy flow is live on the latest testnet0 dashboard. 🔹 Ephemeral private sessions - Current v1 rough flow is also now surfaced on the latest testnet0 dashboard. With testnet0 now updated on the dashboard side, we’ll start running more full E2E privacy tests on the testnet environment from here. 🔹 What’s next - If the latest flow looks stable enough, we’ll keep pushing forward with wider test coverage. - We’re also deploying the latest PrivacySettingData module (v1, covering both ephemeral + dedicated paths) to testnet1 shortly. - Right now, testnet0 is the most up-to-date environment, including the dashboard changes. So the main focus now is simple: run more full-flow tests, refine what breaks, and harden both privacy paths before broader rollout. #Cortensor #DevLog #Privacy #EncryptedInference #Dashboard #DedicatedNodes #EphemeralNodes #Testnet0 #DePIN

🛠️ DevLog – Latest Privacy Dashboard Changes Live on Testnet0 We’ve deployed the latest dashboard privacy changes to testnet0 so it now supports both: 🔹 Dedicated private sessions - Current v0.5 dedicated-node privacy flow is live on the latest testnet0 dashboard. 🔹 Ephemeral private sessions - Current v1 rough flow is also now surfaced on the latest testnet0 dashboard. With testnet0 now updated on the dashboard side, we’ll start running more full E2E privacy tests on the testnet environment from here. 🔹 What’s next - If the latest flow looks stable enough, we’ll keep pushing forward with wider test coverage. - We’re also deploying the latest PrivacySettingData module (v1, covering both ephemeral + dedicated paths) to testnet1 shortly. - Right now, testnet0 is the most up-to-date environment, including the dashboard changes. So the main focus now is simple: run more full-flow tests, refine what breaks, and harden both privacy paths before broader rollout. #Cortensor #DevLog #Privacy #EncryptedInference #Dashboard #DedicatedNodes #EphemeralNodes #Testnet0 #DePIN

🛠️ DevLog – Privacy Flows Working Roughly on Testnet0, More Testing Today At this point, both privacy paths are now working in rough form on testnet0: 🔹 Dedicated private sessions (v0.5) - Router-driven private / encrypted flow is working end to end in rough form. - Dashboard can fetch/cache the key and show decrypted input/output for the authorized owner. 🔹 Ephemeral private sessions (v1.0) - We’ve also now covered the first rough E2E case for ephemeral-node privacy. - The critical code path is there, but it still needs more regression testing and refinement. 🔹 What we’re doing today - Run more full-flow tests on both dedicated and ephemeral privacy paths. - Refine rough edges across router, miner, privacy settings, and dashboard UX. - Keep tightening the owner-side dashboard flow, key handling, and private task/result views. 🔹 Rollout approach - We’ll keep all of this confined to testnet0 for now while we polish the flows and UI further. - Once the dashboard side and regression pass feel cleaner, we’ll push the same privacy flow more fully into testnet1a as well. So the main shift now is from “can this path work at all?” to “can we make both privacy flows stable and usable enough before wider rollout?” Own Your Key. Own Your Data. #Cortensor #DevLog #Privacy #EncryptedInference #DedicatedNodes #EphemeralNodes #Testnet0 #DePIN

🛠️ DevLog – Privacy v1.0: First Rough E2E Case for Ephemeral Sessions We’ve now run the first rough E2E case for privacy v1.0 / ephemeral private sessions. 🔹 What this covered - In this test flow, an ephemeral node was assigned to a private session and successfully requested its encryption key from the router for inference. - That means the critical ephemeral privacy path is now covered at least once: - private ephemeral session - node assignment - router ACL/privacy check - key request from miner - inference path with the private-session flow 🔹 Current status - This is still fragile / early, but it’s an important first milestone for ephemeral privacy. - The dashboard side is also far enough along to reflect private ephemeral session state during this flow. 🔹 What’s next - Keep testing this on testnet0 further. - Run more regression checks and tighten the ephemeral privacy flow before we go any wider. So while it’s still rough, this is the first real E2E case for privacy v1.0 on ephemeral-node sessions. #Cortensor #DevLog #Privacy #EncryptedInference #EphemeralNodes #Testnet0 #DePIN

🛠️ DevLog – Ephemeral Private Session Flow: Dashboard Iteration Progress Follow-up on the earlier privacy v1.0 / ephemeral session work: we’ve now done another iteration and can test some of the dashboard-side flow. 🔹 What’s working so far - We can now create an ephemeral-node session and mark it as private from the dashboard. - The dashboard flow is far enough along that session state, privacy state, and key-related UI can be exercised for ephemeral sessions as well. 🔹 What’s still blocked - We can’t really run the full E2E ephemeral privacy test yet. - For that, we need to upgrade the miners used by node operators so they can follow the newer ephemeral privacy flow correctly. 🔹 What happens next - Finish the dashboard-side iteration first. - Then coordinate the miner upgrade path with node ops and continue testing the full ephemeral private-session flow after that. #Cortensor #DevLog #Privacy #EncryptedInference #EphemeralNodes #Dashboard #DePIN

🛠️ DevLog – Privacy v1.0 / Ephemeral Node Support: Rough Iteration in Progress We’ve started the first rough code iteration for privacy v1.0 / ephemeral private sessions based on the latest draft spec. 🔹 What changed - We made rough changes in the Session module so ephemeral node assignment/removal can eventually work with the privacy flow as well. - We’re also preparing the latest Session and PrivacySettingData modules together so we can test this path as one system. 🔹 Current status - This is still experimental / untested: - no real deployment yet - no end-to-end tests yet - We’ll do another review pass this afternoon before pushing further. 🔹 What happens next - We’ll iterate this on testnet0 first, so some user-task flow may be unstable while we test the new ephemeral privacy path. #Cortensor #DevLog #Privacy #EncryptedInference #EphemeralNodes #Testnet0 #DePIN

🛠️ DevLog – PrivacySettingData Prep for Ephemeral Privacy v1.0 We’ve already made a rough update to PrivacySettingData so it can start moving toward the ephemeral-session privacy design, while still staying compatible with the current dedicated-session flow. 🔹 What changed - PrivacySettingData was updated as an early prep step for Privacy v1.0 / ephemeral private sessions. - The immediate goal is to make sure these changes do not introduce regression for the current dedicated-node privacy flow. 🔹 What happens next - We’ll push this change to testnet0 first and test it there before going wider. - If it behaves cleanly with dedicated private sessions, that gives us a better base to keep shaping the module for ephemeral-session support. 🔹 Where we are on v1.0 - We’re still thinking through what other building blocks are needed for Private Ephemeral Sessions / Privacy v1.0. - This module update is mostly about making sure the privacy data layer is ready early, before the rest of that design settles. #Cortensor #DevLog #Privacy #EncryptedInference #EphemeralNodes #DedicatedNodes #DePIN

🛠️ DevLog – Rough v1.0 Design for Private Ephemeral Sessions We’ve started rough design thinking for privacy feature v1.0, which would extend private / encrypted inference from dedicated-node sessions to ephemeral-node sessions. 🔹 Current direction - Reuse the existing PrivacySettingData module instead of creating a separate privacy data path. - Add a few more mappings so the module can track both: - dedicated private sessions - ephemeral private sessions 🔹 How ephemeral privacy would differ - Unlike dedicated sessions, ephemeral private sessions would need to be managed more dynamically through the Session module. - When a user enables privacy on an ephemeral session, the Session module would: - add assigned ephemeral nodes into PrivacySettingData ACL for encryption-key access - remove them again when those nodes are released from the session - This keeps the privacy flow conceptually similar to today’s v0.5 dedicated-session model, but adapted to nodes that come and go. 🔹 Where we are right now - We’ve done a rough pass over the Session module to identify where those add/remove hooks likely need to happen. - This is still very early design work for v1.0, not implementation yet. Once the dedicated-session privacy flow is more stable, we’ll come back to this, refine the design, and then implement/test the ephemeral-session path. #Cortensor #DevLog #Privacy #EncryptedInference #EphemeralNodes #DedicatedNodes #DePIN

🛠️ DevLog – Rough v1.0 Design for Private Ephemeral Sessions We’ve started rough design thinking for privacy feature v1.0, which would extend private / encrypted inference from dedicated-node sessions to ephemeral-node sessions. 🔹 Current direction - Reuse the existing PrivacySettingData module instead of creating a separate privacy data path. - Add a few more mappings so the module can track both: - dedicated private sessions - ephemeral private sessions 🔹 How ephemeral privacy would differ - Unlike dedicated sessions, ephemeral private sessions would need to be managed more dynamically through the Session module. - When a user enables privacy on an ephemeral session, the Session module would: - add assigned ephemeral nodes into PrivacySettingData ACL for encryption-key access - remove them again when those nodes are released from the session - This keeps the privacy flow conceptually similar to today’s v0.5 dedicated-session model, but adapted to nodes that come and go. 🔹 Where we are right now - We’ve done a rough pass over the Session module to identify where those add/remove hooks likely need to happen. - This is still very early design work for v1.0, not implementation yet. Once the dedicated-session privacy flow is more stable, we’ll come back to this, refine the design, and then implement/test the ephemeral-session path. #Cortensor #DevLog #Privacy #EncryptedInference #EphemeralNodes #DedicatedNodes #DePIN

🛠️ DevLog – Thinking Ahead on Ephemeral Privacy (v1.0) While we keep running more tests on privacy feature v0.5 for dedicated sessions, we’re also starting to think ahead about privacy v1.0 for ephemeral-node sessions. 🔹 Current direction - Instead of introducing a completely new data module, we’ll first check whether we can reuse and extend the current privacy settings module for ephemeral private-session flow. - The goal is to keep the privacy model more unified instead of splitting dedicated and ephemeral paths too early. 🔹 Why ephemeral privacy is trickier - Nodes in ephemeral sessions come and go, so the privacy ACL has to keep up with that lifecycle. - That means the system likely needs to: - add nodes into the privacy settings when they get assigned to a session - remove them when they leave - Two issues already stand out: - current node-pool selection flow is already a bit crowded and may need cleanup first - timing matters, because encryption-key access must be available exactly when the assigned node needs it 🔹 What this means - This is still high-level design thinking, not implementation yet. - While v0.5 gets more regression testing, we’ll use this time to scope out what a privacy v1.0 design for ephemeral sessions should look like. #Cortensor #DevLog #Privacy #EncryptedInference #EphemeralNodes #DedicatedNodes #DePIN

🛠️ DevLog – Thinking Ahead on Ephemeral Privacy (v1.0) While we keep running more tests on privacy feature v0.5 for dedicated sessions, we’re also starting to think ahead about privacy v1.0 for ephemeral-node sessions. 🔹 Current direction - Instead of introducing a completely new data module, we’ll first check whether we can reuse and extend the current privacy settings module for ephemeral private-session flow. - The goal is to keep the privacy model more unified instead of splitting dedicated and ephemeral paths too early. 🔹 Why ephemeral privacy is trickier - Nodes in ephemeral sessions come and go, so the privacy ACL has to keep up with that lifecycle. - That means the system likely needs to: - add nodes into the privacy settings when they get assigned to a session - remove them when they leave - Two issues already stand out: - current node-pool selection flow is already a bit crowded and may need cleanup first - timing matters, because encryption-key access must be available exactly when the assigned node needs it 🔹 What this means - This is still high-level design thinking, not implementation yet. - While v0.5 gets more regression testing, we’ll use this time to scope out what a privacy v1.0 design for ephemeral sessions should look like. #Cortensor #DevLog #Privacy #EncryptedInference #EphemeralNodes #DedicatedNodes #DePIN

🛠️ DevLog – Private Dedicated Session Flow Working on Testnet0/Testnet1a, Next Up: Regression & Private Ephemeral Design With the rough E2E private-session for dedicated node flow now working on both testnet0 and testnet1a, the next step is to keep tightening it through more regression checks and UX iteration. 🔹 What’s next on the current flow - Run more regression tests around the privacy path on both environments. - Keep refining any rough UX edges around session privacy toggle, key fetch/cache, and decrypted input/output views for the owner. 🔹 What we’ll think through in parallel - Start rough design work for ephemeral-node session privacy as well. - This path will need more adjustment than dedicated sessions, since ephemeral nodes come and go and privacy settings / ACL flow will need to adapt to that lifecycle. - So for now, the plan is to think through the design first before pushing into implementation. So the dedicated-session flow is now far enough along to keep hardening, while ephemeral-node privacy becomes the next design problem to shape in parallel. #Cortensor #DevLog #Privacy #EncryptedInference #DedicatedNodes #EphemeralNodes #Testnet0 #Testnet1a #DePIN