Skip to content

0.2.0 - 2026-07-13

Choose a tag to compare

View all tags
@github-actions github-actions released this 15 Jul 06:55
v0.2.0
b972abf
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Release Notes

Agents, edge controls, and a safety-critical rewrite of the whole binary. riz
is now an agent2agent-protocol server in its own right, gates the data plane
per caller, and every line compiled into the binary is held to NASA's Power of
10 rules.

Added

  • A2A built-in agent — set [agent] and this instance becomes an
    agent2agent-protocol server: an Agent Card at
    /.well-known/agent-card.json and a JSON-RPC endpoint at /_riz/a2a where
    peers delegate tasks. It reasons through the LLM gateway with this instance's
    own functions as tools, streams live task events over SSE
    (SendStreamingMessage), and forms a mesh via [agent.peers] with
    hop-capped delegation (riz a2a send).
  • Per-caller API keys + token-bucket rate limiting — [api_keys.<name>]
    maps a caller to a secret + rate ceiling. Non-/_riz/* requests (function
    invocations, WebSocket handshakes, colocated static) must present a matching
    X-Api-Key; unknown/absent keys fail closed (401) and each caller has its
    own bucket (429 + Retry-After). No keys → open, unchanged.
  • Structured audit log — deploy, config-reload, and auth-denial events on
    the riz.audit tracing target, scrubbed of secret material; route them with
    RUST_LOG=riz.audit=info.
  • Production metrics — the four golden signals plus worker supervision and
    cache efficiency at /_riz/metrics, including saturation
    (riz_concurrency_in_use/_limit, riz_admission_rejected_total) and a
    cross-instance-aggregatable riz_request_duration_seconds histogram. A
    /_riz/ready readiness probe; [metrics] enabled off switch. See
    docs/METRICS.md.
  • LLM gateway upgrades — OpenAI function-calling across every provider,
    token-level streaming passthrough for OpenAI-compatible upstreams, and
    Anthropic-native streaming translated to OpenAI chunks on the fly.
  • MCP — resources (a live instance describes itself to agents) and
    WebSocket functions exposed as tools via ephemeral sessions.
  • --dev TUI — live log search (/) + severity filter (l), a ? help
    overlay, an Enter-to-open invocation inspector (recent calls per function),
    and a saturation column in the Processes tab.
  • Runtimes & scaffolding — per-function environment variables
    ([function.<name>.env]); a wasm-http init template (a wasm32-wasip1
    handler in the WASI sandbox); the full-stack ai-chat template (React chat
    UI + a server-side agent loop through the gateway).
  • Per-worker seccomp-BPF — a deny-EPERM blocklist of 22 escape/tamper
    syscalls in pre_exec, stacked on rlimits + prctl + Landlock.
  • Actionable startup errors — bad riz.toml (points at the field), missing
    runtime binary (names it + install hint), and port-in-use (names the port +
    how to change it) now say how to fix themselves.
  • CI throughput floor — a conservative, non-flaky regression tripwire for
    HTTP dispatch (the 91k req/s headline stays a wrk bench recipe).

Changed

  • Safety-critical posture — NASA's Power of 10, adapted to Rust
    (docs/SAFETY.md), is now binding for everything compiled into the binary:
    no unwrap/expect/panic/indexing/unchecked-arithmetic on runtime data,
    bounded channels, supervised loops. 253 flagged sites were driven to zero and
    the lints promoted to a three-tier enforced gate (workspace deny + a
    --lib --bins CI gate + a ratchet that only decreases).
  • Supply chain — a cargo-deny CI gate, a CycloneDX SBOM, and keyless
    GitHub build-provenance attestations (SLSA via Actions OIDC) on every release
    artifact.
  • Static serving streams file bodies (flat per-connection memory, no HEAD
    reads).
  • Production hardening (Phase 1) — JWKS authorizer cache, reflected-origin
    credentialed-CORS rejection, and a hot-swap pool rebuild that resizes
    admission.

Fixed

  • WebSocket function routes now pass the per-caller API-key gate (they
    previously bypassed it via explicit route mounting); keyed requests bypass
    the response cache (no cross-caller serving); a startup warning fires when
    [api_keys] is set but [auth] bearer_token is unset (the /_riz/* plane,
    including MCP tool-calls, stays open otherwise).
  • MCP no longer advertises WebSocket functions as directly callable tools.

Install riz 0.2.0

Install prebuilt binaries via shell script

curl --proto '=https' --tlsv1.2 -LsSf https://github.com/24X7/riz/releases/download/v0.2.0/riz-installer.sh | sh

Download riz 0.2.0

File Platform Checksum
riz-aarch64-apple-darwin.tar.xz Apple Silicon macOS checksum
riz-aarch64-unknown-linux-gnu.tar.xz ARM64 Linux checksum
riz-x86_64-unknown-linux-gnu.tar.xz x64 Linux checksum

Verifying GitHub Artifact Attestations

The artifacts in this release have attestations generated with GitHub Artifact Attestations. These can be verified by using the GitHub CLI:

gh attestation verify <file-path of downloaded artifact> --repo 24X7/riz

You can also download the attestation from GitHub and verify against that directly:

gh attestation verify <file-path of downloaded artifact> --bundle <file-path of downloaded attestation>
Assets 12
Loading