Add SSE streaming for AI responses #3

New Issue

hermes · 2026-03-18T22:01:00Z

hermes commented

2026-03-18 22:01:00 +00:00

Context

Currently the agent returns complete text after execution finishes. For a good
Workshop UX, responses should stream token-by-token as Claude generates them.

Requirements

Add SSE (Server-Sent Events) streaming to the work execution phase
When a job enters executing state, stream partial results via SSE
Endpoint: GET /api/jobs/:id/stream (returns text/event-stream)
Also support streaming for Session Mode 2 requests
Use Anthropic's streaming API (stream: true on messages.create)

Event Format

data: {"type": "token", "content": "The"}
data: {"type": "token", "content": " Lightning"}
data: {"type": "token", "content": " Network"}
data: {"type": "done", "result": "The Lightning Network is...", "inputTokens": 50, "outputTokens": 120}

Acceptance Criteria

SSE endpoint works for Mode 1 jobs
SSE endpoint works for Mode 2 session requests
Tokens appear within 200ms of generation
Final event includes complete result + token counts
Connection properly closes on completion or error
Backpressure handling (slow clients)

References

Parent epic: #{epic_num}

## Context Currently the agent returns complete text after execution finishes. For a good Workshop UX, responses should stream token-by-token as Claude generates them. ## Requirements - Add SSE (Server-Sent Events) streaming to the work execution phase - When a job enters `executing` state, stream partial results via SSE - Endpoint: `GET /api/jobs/:id/stream` (returns `text/event-stream`) - Also support streaming for Session Mode 2 requests - Use Anthropic's streaming API (`stream: true` on messages.create) ## Event Format ``` data: {"type": "token", "content": "The"} data: {"type": "token", "content": " Lightning"} data: {"type": "token", "content": " Network"} data: {"type": "done", "result": "The Lightning Network is...", "inputTokens": 50, "outputTokens": 120} ``` ## Acceptance Criteria - [ ] SSE endpoint works for Mode 1 jobs - [ ] SSE endpoint works for Mode 2 session requests - [ ] Tokens appear within 200ms of generation - [ ] Final event includes complete result + token counts - [ ] Connection properly closes on completion or error - [ ] Backpressure handling (slow clients) ## References - Parent epic: #{epic_num}

hermes added the api enhancement replit labels 2026-03-18 22:01:00 +00:00

replit self-assigned this 2026-03-18 22:02:21 +00:00

replit referenced this issue

2026-03-18 22:37:59 +00:00

feat: streaming, rate limiting, CORS, event bus, SSE — closes #2 #3 #4 #5; fixes #14 #16 #20

replit referenced this issue from a commit

2026-03-19 00:43:03 +00:00

feat: WebGL context loss recovery for iPad PWA (#14)

replit referenced this issue from a commit

2026-03-19 00:43:03 +00:00

feat: WebGL context loss recovery for iPad PWA (#14)

replit referenced this issue