v0.1.79 — Build → Eval → Run → AgentCert → Deploy → Publish, with a real TrustScore at every beat

Released 2026-09-26. Signed + notarized on macOS (Apple Silicon + Intel); Windows + Linux published. Gated on full smoke + E2E journeys.

This release makes the whole TrustModel story demo-able end to end — from a brand-new agent to a public certificate — and puts it where you can actually see it: on the agent’s lifecycle rail.

The full lifecycle is on the left rail now

The Agents panel used to hide everything but Build until an agent existed — so you couldn’t find Run, Eval, or AgentCert when you needed them. The rail now shows the whole lifecycle, in order:

Build → Eval → Run → AgentCert → Deploy → Publish

Every stage is visible right after Build (dimmed until your first agent exists, never hidden), so the path from “I just built an agent” to “it has a public certificate” reads as one legible spine.

A real TrustScore — from one API key

Paste a TrustModel Starter API key once (AgentCert → Add API key, which opens the console for you), and AgentFoundry calls the live TrustModel /sdk/v1 API directly:

• Eval runs a scored evaluation → a real TrustScore (0–100) across 11 governance dimensions (safety, security, reliability, accuracy, transparency, accountability, privacy, fairness, robustness, compliance, human oversight).
• AgentCert mints a public certificate with a verify.trustmodel.ai verification URL — the shareable proof your agent was evaluated.

No Edge sidecar, no policy to publish, no HMAC setup — just a key. And no mocks: a TrustScore or certificate is only ever shown from a real TrustModel response. If scoring can’t complete, you get an honest error, never a fabricated number.

The score shows up where you work

• Eval — a TrustScore strip with one-tap Run TrustModel Eval.
• Run — after you fire a prompt, Score this run scores the exact output the agent just produced.
• AgentCert — the full gauge, all 11 dimension bars, and the public certificate card with its verify link.

All three read the same stored result, so the number is consistent across the rail — and it carries into Business mode.

A non-coder can raise the score

This is the part that makes it real for the people who oversee agents but don’t write code. In Business mode, a PM or ops owner sees the same TrustScore on the Fleet badge and in the Guardrails panel. Because skills and guardrails are injected as the agent’s system prompt, they reshape what the agent actually says — so:

1. Open Skillify, run the agent on a scenario, hit Score with TrustModel.
2. Add a skill or tighten a guardrail (data scopes, allowed actions, human-approval).
3. Score the new response — and watch a before → after delta as the number climbs.

It’s a genuine loop: the score moves because the agent’s real output got better, not because a slider did.

Upgrade

• Desktop: update to v0.1.79 (auto-updates on relaunch). macOS arm64 + x64, Windows, Linux AppImage.
• Bring a TrustModel Starter key: create a free one at app.trustmodel.ai  → Settings → API Keys. Everything else — the rail, the strips, the Business-mode loop — is already in the build.

Known issue: the Linux AppImage may not launch on very old distributions (e.g. Ubuntu 20.04) due to a glibc mismatch in a native module; a broader-compatibility Linux build is in progress. macOS and Windows are unaffected.