Pistachio v0.30: SSE Streaming, Throughput Benchmarks, and Readiness Diagnostics

Pistachio v0.30. Chat completions now support SSE with provenance metadata on every response. A built-in benchmark command measures throughput against any gateway. New `pistachio doctor` command checks wallet, RPC, balances, passes, sessions, and providers in one shot.

SSE streaming and response metadata

`/v1/chat/completions` now accepts `stream: true` and responds with Server-Sent Events. The SSE protocol layer is complete: role, content, thinking, and finish_reason events, followed by a `_pistachio` metadata event and `[DONE]`. Enable streaming in Chat Settings to try it.

Current Morpheus providers return the entire completion as a single response, not per-token. The text appears all at once. When providers add per-token TCP delivery, Pistachio will stream incrementally with no API change on the client side.

A canonical response JSON is appended as a `_pistachio` SSE event before `[DONE]`, carrying session ID, provider address, MOR stake, token usage, and latency. This is the same provenance metadata that non-streaming responses include inline. Streaming and non-streaming cards display identical detail.

Benchmark command

`pistachio benchmark` is a new CLI tool for measuring provider throughput. Point it at any Pistachio gateway, pick a model, set concurrency, and get per-request latency, tokens per second, success rate, and error classification. It uses the same HTTP path as normal inference, so it tests the full stack: auth, session selection, provider communication, and response parsing.

Three prompt tiers (tiny, medium, large) exercise different payload sizes. Concurrent lanes test how many parallel requests a provider can actually handle versus what the marketplace advertises. Output as JSON for programmatic analysis or as a formatted report.

DRM3 runs throughput audits against the Morpheus network daily using this tool. Results are published in the repo. If you are evaluating a provider or comparing models, `pistachio benchmark` gives you the numbers before you commit stake.

pistachio benchmark --url https://gateway.example --api-key pst_... --model llama-3.3-70b --max-tokens 512

pistachio benchmark --url https://gateway.example --api-key pst_... --json > results.json

Readiness diagnostic

`pistachio doctor` checks everything required for inference in one command: wallet loaded, RPC reachable, ETH balance for gas, MOR balance for staking, DRM3 Pass capacity, active sessions, and reachable providers. Each check reports pass or fail with actionable context. Run it after a fresh install or when something is not working.

pistachio doctor

Auto-reclaim and Core Models

Auto-reclaim is now a toggle in Settings. When enabled, Pistachio automatically closes expired sessions and returns the locked MOR to your wallet. The sidebar shows reclaim status. Previously, expired sessions could sit indefinitely, locking MOR that could be restaked.

A new "Core Models" stock template ships with 13 models at 7 MOR each, balanced fill. It provides a working starting configuration covering the most reliable models on the network.

Gateway: RPC endpoint hiding

Gateway operators can now hide RPC endpoint URLs from the Settings page. Set `HIDE_RPC_ENDPOINTS=true` and any URL containing an API key is redacted in the UI and API responses. Set `HIDE_RPC_MESSAGE` to show a custom banner explaining that RPC configuration is managed by the operator. Useful for shared gateways where you do not want users seeing your Alchemy keys.

Reclaim reliability

Two fixes improve session reclaim reliability. First, zombie detection now runs on individual operation lookups, not just the active operations list. Previously, a stuck reclaim operation could survive indefinitely because the UI polled by operation ID, which bypassed the garbage collector. Second, reclaim operations are now panic-safe. If the reclaim task panics, the operation is marked Failed instead of staying in Running state forever.

Install or upgrade

Desktop app users will receive an auto-update prompt on next launch. Full documentation at drm3.io/products/pistachio. Release notes and binaries at github.com/drm3labs/drm3-releases.

brew tap drm3labs/drm3 && brew install pistachio

brew tap drm3labs/drm3 && brew install --cask pistachio-desktop

curl -fsSL https://drm3.network/install.sh | sh

docker pull drm3/pistachio:latest