From 457d9f0c9736c420d1651500159c0aa93b88ff35 Mon Sep 17 00:00:00 2001 From: sim Date: Thu, 17 Sep 2026 18:34:05 +0100 Subject: [PATCH] docs(readme): v0.9.5 release banner, DigitalOcean install, refreshed features MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit - Release banner β†’ v0.9.5, linking the What's New overview and the full notes; the v0.9.0 review video stays as the previous release's link. - Quick Start: first-time setup describes the first-run sequence (Connect Claude is the one required step), and adds the guided DigitalOcean install and the prebuilt-image (`start.sh --hosted`) path. - The create-agent wizard list matches create-agent 2.0 (the domain wizards were retired). - Features: Workspace (chats as tabs, rail, live cards, asks, voice mode), Canvas, subscription usage, machine identities, first-run setup, prebuilt images, opt-in telemetry; the model list includes Fable. Co-Authored-By: Claude Opus 5 (1M context) --- README.md | 53 +++++++++++++++++++++++++++++++++++++++++------------ 1 file changed, 41 insertions(+), 12 deletions(-) diff --git a/README.md b/README.md index b5b320c03f..a2dfd79e2b 100644 --- a/README.md +++ b/README.md @@ -24,7 +24,22 @@

> [!NOTE] -> **πŸ“¦ Just released β€” [v0.9.0](docs/releases/0.9.0.md)** (August 17, 2026): **Workspace** β€” the client-facing surface, now in every build β€” plus fresh-install provisioning (remote template registry, GitHub import wizard, manifest install UI, guided credentials), the multi-source skills library and Library page, bidirectional A2A, dashboard List view + grid data tiles, automatic database backups, and a large reliability & security sweep. [Watch the release review β†’](https://youtu.be/fia2qa4YcLg) Β· [What's new overview β†’](https://docs.ability.ai/whats-new/v0.9.0) Β· [Full release notes β†’](docs/releases/0.9.0.md) +> **πŸ“¦ Just released: [v0.9.5](docs/releases/0.9.5.md)** (September 17, 2026). **Workspace** is where the work happens now: +> - chats per topic +> - a conversation rail +> - live execution cards +> - agents that ask you questions +> - a canvas for every agent +> - real-time **voice mode** +> +> Also new: +> - prebuilt images and a guided **[DigitalOcean install](https://docs.ability.ai/getting-started/deploying/digitalocean)** +> - first-run setup in the browser +> - subscription usage and headroom you can see +> - opt-in instance telemetry +> - credentials encrypted at rest +> +> [What's new overview β†’](https://docs.ability.ai/whats-new/v0.9.5) Β· [Full release notes β†’](docs/releases/0.9.5.md) Β· Previous: [v0.9.0 release review video β†’](https://youtu.be/fia2qa4YcLg)

@@ -158,10 +173,9 @@ For a fully non-interactive bring-up, run `./scripts/deploy/start.sh --unattende **First-time setup** -1. Open http://localhost β€” you'll be redirected to the setup wizard -2. Enter your **admin email** β€” this becomes your sign-in identity β€” and an **admin password** (minimum 12 characters, with upper- and lowercase, a digit, and a special character) -3. Log in with that email (or the username `admin`) and your password -4. Go to **Settings** β†’ **Integrations** β†’ **API Keys** to configure your Anthropic API key +1. Open http://localhost and log in with the username `admin` and the `ADMIN_PASSWORD` from `.env`. +2. **First-run setup** opens on the Dashboard. The one required step is **Connect Claude**: paste a Claude subscription token or an Anthropic API key, and Trinity checks it before saving. +3. Optional steps follow: GitHub, email-provider and Gemini keys, then your first agent. Anything you skip is under **Settings β†’ Integrations** later. **Access** @@ -171,6 +185,10 @@ For a fully non-interactive bring-up, run `./scripts/deploy/start.sh --unattende > **Don't want to self-host?** Trinity also runs as a managed instance on any cloud you control. [Talk to an engineer β†’](mailto:hello@ability.ai) β€” an engineer reads this, not a CRM. Reply in one business day, your time zone. +> **On DigitalOcean?** The guided installer (`scripts/deploy/trinity-do-create.sh`) takes a few steps and about ten minutes. You install `doctl`, create a Claude subscription token, choose an admin password, and run one command. The result is Trinity on its own Droplet behind HTTPS. See [Deploy on DigitalOcean](https://docs.ability.ai/getting-started/deploying/digitalocean). + +> **Prebuilt images:** `./scripts/deploy/start.sh --hosted` pulls the published GHCR images instead of building from source. Pin `TRINITY_IMAGE_TAG` (for example `v0.9.5`) on any instance you keep. + > **Deploying to a remote server?** `/trinity:deploy-new-instance` from the [abilities marketplace](https://github.com/abilityai/abilities) provisions Trinity on any server you can SSH into β€” and scaffolds the [Trinity Ops Agent](https://github.com/abilityai/trinity-ops-public) to manage it (health, logs, updates, rollback, and SQLiteβ†’PostgreSQL migration). ### Phase B β€” Build & deploy agents from Claude Code @@ -190,8 +208,8 @@ Most of the Trinity workflow lives in the **[abilities plugin marketplace](https # 3. Scaffold an agent β€” or start from any existing Claude Code agent directory /create-agent:create -# β†’ Pick a wizard (prospector, chief-of-staff, recon, ghostwriter, kb-agent, …) -# or /create-agent:custom for a blank canvas +# β†’ /create-agent:custom scaffolds any agent from scratch; +# /create-agent:website and /create-agent:clone cover the other paths # 4. Deploy it to Trinity /trinity:onboard @@ -279,6 +297,7 @@ The full feature set is below and in the [documentation](#documentation). - **List View** β€” The fleet as a sortable, filterable roster with inline run/autonomy toggles; `/` type-to-filter works across all dashboard views - **Grid Data Tiles** β€” Fleet info tiles alongside agent tiles (executions by trigger, recent failures), an org overlay of departments and reporting lines - **Timeline View** β€” Gantt-style execution timeline with trigger-based color coding (manual, scheduled, MCP, agent-triggered, public, paid) +- **Subscription Usage** β€” 5-hour and 7-day usage per Claude subscription with a per-agent breakdown and rate-limit history, a Subscription pressure tile, headroom history, and an alert at 75% of the weekly limit - **Host Telemetry** β€” Real-time CPU, memory, and disk monitoring in the dashboard header - **Fleet Health Monitoring** β€” Multi-layer health checks (Docker, network, business) with alerting and WebSocket updates - **OpenTelemetry Metrics & Tracing** β€” Cost, token usage, and productivity tracking exportable to Grafana/Datadog; distributed traces across multi-agent calls @@ -287,7 +306,7 @@ The full feature set is below and in the [documentation](#documentation). - **Isolated Docker Containers** β€” Each agent runs in its own container with dedicated resources - **Multi-Runtime Support** β€” Choose between Claude Code (Anthropic), OpenAI Codex, or Gemini CLI (Google) per agent -- **Model Selection** β€” Choose which Claude model (Opus, Sonnet, Haiku) per task or schedule +- **Model Selection** β€” Choose which Claude model (Fable, Opus, Sonnet, Haiku) per task or schedule - **Agent Overview** β€” Per-agent Overview tab with activity trends, success rate, duration, and health over a 7/14/30-day window - **Compatibility Report** β€” Advisory validation of a running agent's workspace against 88 best-practice checks in 12 categories, with one-click fixes for common issues - **Agent Dashboard** β€” Custom dashboards defined via `dashboard.yaml` with 11 widget types, historical tracking, and sparkline visualization @@ -317,7 +336,14 @@ The full feature set is below and in the [documentation](#documentation). - **Voice** β€” Spoken voice replies (text-to-speech) on Slack, Telegram, and WhatsApp, plus opt-in outbound phone calls over Twilio telephony - **Unified Access Control** β€” Verified-email allow-list governs access across web, Slack, and Telegram with per-agent `require_email` / `open_access` policies (#311) - **Proactive Messaging** β€” Agents initiate user conversations via `send_message` / `send_group_message` MCP tools (#321, #349) -- **Workspace** β€” A client-facing surface at `/workspace` for the people your agents serve: verified-email sign-in, chats with one or more agents, agent pages, skill hints, streaming turns, file uploads β€” one click away for platform users +- **Workspace** β€” Where people work with agents, at `/workspace`, with verified-email sign-in for the people your agents serve: + - several chats per agent as tabs, and multi-agent rooms + - a conversation rail with Info, Loops, Files and Canvas tabs + - live execution cards in the chat, and loops you start and stop from the chat + - agents that ask a named person a question + - deliverables and ratings, and a canvas for every agent + - real-time **voice mode** over the current chat +- **Canvas** β€” A rich block vocabulary (charts, images, diagrams) rendered the same in chat and during a call, with a design kit, read-only share links, PDF export, and the open canvas as shared context for the agent - **Agent Reports** β€” Agents publish structured reports (KPI, table, markdown, timeline) rendered on the dashboard, exportable to Excel/PDF, searchable, and readable back over MCP ### Operations @@ -329,6 +355,7 @@ The full feature set is below and in the [documentation](#documentation). - **Subscription Management** β€” Centralized Claude Max/Pro subscription tokens shared across agents - **Platform Audit Log** β€” Append-only cross-cutting audit trail for lifecycle, auth, and MCP events (SEC-001, admin-only API) - **Role Hierarchy** β€” 4-tier RBAC (`user` < `operator` < `creator` < `admin`) with whitelist-driven role on first login +- **Machine Identities** β€” Service credentials for admin and ops APIs that keep working when two-factor sign-in is enforced for people - **Agent Tags & System Views** β€” Organize agents with tags and saved filter views for fleet management - **Live Execution Streaming** β€” Real-time streaming of execution logs to the web UI - **Execution Termination** β€” Stop running executions gracefully via SIGINT/SIGKILL @@ -340,8 +367,10 @@ The full feature set is below and in the [documentation](#documentation). - **Public Agent Links** β€” Shareable links for unauthenticated agent access with session persistence and Slack integration - **Paid Agent Access (x402)** β€” Per-agent monetization via Nevermined x402 payment protocol - **Mobile Admin PWA** β€” Standalone mobile admin at `/m`, installable as a home screen app -- **First-Time Setup Wizard** β€” Guided setup for admin password and API key configuration +- **First-Run Setup** β€” A browser setup sequence on the Dashboard: connect Claude (checked before saving), GitHub, email and Gemini keys, and a first agent, all without a terminal. A Marketplace droplet's admin is claimed in the browser. +- **Prebuilt Images & DigitalOcean Install** β€” Published GHCR images (`start.sh --hosted`), an unattended cloud-init first boot, and a guided DigitalOcean installer - **Automatic Database Backups** β€” Nightly verified recovery points for SQLite and PostgreSQL, plus a pre-migration copy at boot +- **Opt-in Instance Telemetry** β€” Anonymised operational signal, sent only after an admin consents, with a send log showing where each share went ## Architecture @@ -503,7 +532,7 @@ Abilities structures agent work as a four-step lifecycle: /agent-dev:add-backlog (or: trinity deploy .) /create-agent:adjust ``` -**Scaffold** β€” wizards like `/create-agent:prospector` or `/create-agent:custom` produce a fully configured, Trinity-compatible agent: `CLAUDE.md`, initial skills, `template.yaml`, a metrics dashboard, and an onboarding tracker. +**Scaffold** β€” `/create-agent:custom` (or `/create-agent:website`) produces a fully configured, Trinity-compatible agent: `CLAUDE.md`, initial skills, `template.yaml`, a metrics dashboard, and an onboarding tracker. **Develop** β€” `/agent-dev:create-playbook` adds capabilities, `/agent-dev:add-memory` adds persistence (file index, knowledge graph, JSON state, or workspace tracking), `/agent-dev:add-backlog` wires a GitHub Issues workflow. @@ -515,7 +544,7 @@ The 5 plugins: | Plugin | What it does | |--------|-------------| -| **create-agent** | Agent scaffolding: a `/create-agent:create` discovery entry point + 12 wizards (prospector, chief-of-staff, webmaster, recon, receptionist, ghostwriter, kb-agent, doctor, website, custom, clone, adjust) | +| **create-agent** | Agent scaffolding: a `/create-agent:create` entry point, the `custom` and `website` scaffolds, `clone`, and `review` / `adjust` for improving an existing agent | | **agent-dev** | Extend existing agents: playbooks, memory systems, git-sync hooks, GitHub backlog workflow, grooming, sprints, agent-owned pipelines, autonomous work loops | | **trinity** | The platform workflow: `connect`, `onboard`, `sync`, `loop`, `create-dashboard`, `deploy-new-instance` | | **dev-methodology** | Documentation-driven development for any codebase: implementation, testing, security audits, PR validation, release, architecture/schema/config validation |