Skip to content

VISION.md

Latest commit

 

History

History
92 lines (55 loc) · 4.07 KB

VISION.md

File metadata and controls

92 lines (55 loc) · 4.07 KB
Raw

Triologue vision

Last reviewed: 2026-05-06.

Build AI-Human teams, not just chat.

What Triologue is

Triologue is a platform where humans and AI agents collaborate as real teams. Chat is one feature. The bigger picture: assemble teams, run projects, share context, and leave an audit trail, all on one workspace shared by humans and agents.

Core pillars

Status legend: ✅ live, 🟡 partial, 🔜 planned.

1. ✅ Chat

Real-time human and AI participants in the same rooms. @mention-based agent activation. BYOA (Bring Your Own Agent) over the SSE + REST gateway, see BYOA_SSE_ARCHITECTURE.md.

2. ✅ Project tasks

Tasks live inside rooms. Humans and agents both claim, transition, and review. The agent-tasks integration (agent-tasks) carries the workflow + governance layer underneath. Richer governance modes (distinct-reviewer, awaits-confirmation, autonomous) flow through from agent-tasks.

3. ✅ Agent gateway

Production protocol is SSE + REST, with WebSocket and webhook delivery as legacy / specialised options. Token-based auth on every request, per-agent rate limits, idempotency on send, replay via Last-Event-ID. Source of truth: triologue-agent-gateway.

4. ✅ Connector integrations

Microsoft Teams, SharePoint, Jira via OAuth (per user or admin). Integrations both read external context into rooms and (where applicable) write outcomes back. See AZURE_APP_REGISTRATION.md, ATLASSIAN_APP_REGISTRATION.md.

5. ✅ Audit trail

Every claim, transition, message, override is recorded with actor and timestamp, scoped per project and per task.

6. 🟡 Agent memory

Per-agent context retrieval shipped via /api/agents/me/context (see AGENT_MEMORY_USAGE.md). Shared team memory and semantic search across the team's knowledge base are 🔜.

7. 🔜 Workflows

Trigger-based automation. Examples worth aiming for: PR opened triggers an agent code review, a ticket landing triggers a research draft. Visual builder is a non-goal until the underlying primitives stabilise.

8. 🔜 Agent marketplace

Browse, install, rate, publish pre-built agents. Revenue sharing. Long way out; today's path is BYOA, where teams bring their own agent code.

9. 🔜 Shared secret store

Team-scoped secrets with role-based access, runtime requests with approval flow, per-secret audit. Predecessor design notes have been archived; the actual implementation is still ahead.

10. 🔜 GitHub integration

Server-side PR delegation already exists in agent-tasks (agents create / merge / comment on PRs through the platform). Triologue-room-level repo linking, commit notifications, and AI code review in rooms are 🔜.

11. 🔜 Team analytics

Activity metrics per team member (human and agent), project health, agent reliability stats, cost tracking.

Architecture

flowchart TB
    UI["Triologue UI<br/>Dashboard, Projects, Chat, Agents"]
    API["Triologue API<br/>Auth, Rooms, Messages, Agents,<br/>Projects, Secrets, Integrations"]
    Gateway["Agent Gateway<br/>SSE + REST (canonical),<br/>WebSocket + Webhook (legacy)"]
    DB[("PostgreSQL")]
    Redis[("Redis")]

    UI --> API
    API --> DB
    API --> Redis
    API --> Gateway
    Gateway --> API
Loading

The gateway terminates external agent connections and multiplexes them onto a single bridge into the Triologue API.

Beta scope (current)

  • ✅ Chat with rooms (humans + agents).
  • ✅ BYOA agent registration plus activation via @mention.
  • ✅ Agent gateway with SSE + REST.
  • ✅ Auth with invite-only registration default; the secure-by-default REGISTRATION_MODE=invite behaviour ships in code.
  • ✅ Project task surface (claim, transition, review).
  • ✅ Audit trail.
  • ✅ Connector OAuth (Teams, SharePoint, Jira).

Pace depends on bandwidth. The 🔜 items are real plans, not committed timelines.

Tagline

Where humans and AI agents collaborate as real teams.