feat(m2): LLM chat via Ollama — streaming responses with migration context #3

Open

pyr0ball wants to merge 6 commits from feat/m2-llm-chat into feat/m1-system-presence

pyr0ball commented

2026-05-19 07:29:24 -07:00

Owner

Summary

New llm.rs module: build_system_prompt() (migration context baked into system prompt) + chat_stream() (Ollama NDJSON streaming via reqwest::Response::chunk())
New chat Tauri command: builds message list from config, spawns background stream, emits robin:chat-token / robin:chat-done / robin:chat-error
ChatPanel.vue: streaming-aware send() accumulates tokens into pre-inserted message by index; thinking state disables input while Robin is responding; error fallback if Ollama is unreachable

Test plan

Build: npm run build (TypeScript clean)
On Xander's CachyOS with Ollama running (ollama serve, model llama3.2 pulled): open Robin, type a question, verify tokens stream in
With Ollama stopped: verify error message appears in chat
Send a second message after first completes — verify thinking state resets
Open panel, ask question, close panel mid-response — verify no crash on re-open

Notes

Chat is single-turn for M2 (system prompt + current message only) — multi-turn memory is M4
Ollama URL and model configurable in ~/.config/robin/config.toml (defaults: http://localhost:11434, llama3.2)
Base branch is feat/m1-system-presence — will rebase onto main after M1 merges

## Summary - New `llm.rs` module: `build_system_prompt()` (migration context baked into system prompt) + `chat_stream()` (Ollama NDJSON streaming via `reqwest::Response::chunk()`) - New `chat` Tauri command: builds message list from config, spawns background stream, emits `robin:chat-token` / `robin:chat-done` / `robin:chat-error` - `ChatPanel.vue`: streaming-aware `send()` accumulates tokens into pre-inserted message by index; `thinking` state disables input while Robin is responding; error fallback if Ollama is unreachable ## Test plan - [ ] Build: `npm run build` (TypeScript clean) - [ ] On Xander's CachyOS with Ollama running (`ollama serve`, model `llama3.2` pulled): open Robin, type a question, verify tokens stream in - [ ] With Ollama stopped: verify error message appears in chat - [ ] Send a second message after first completes — verify thinking state resets - [ ] Open panel, ask question, close panel mid-response — verify no crash on re-open ## Notes - Chat is single-turn for M2 (system prompt + current message only) — multi-turn memory is M4 - Ollama URL and model configurable in `~/.config/robin/config.toml` (defaults: `http://localhost:11434`, `llama3.2`) - Base branch is `feat/m1-system-presence` — will rebase onto main after M1 merges

pyr0ball added 6 commits 2026-05-19 07:29:24 -07:00

chore(m2): add reqwest to Cargo dependencies 9c45015052

feat(m2): add llm module — build_system_prompt and Ollama streaming client c149c598ae

fix(m2): log emit errors, use Vec<u8> buffer for UTF-8 safety, improve error context 369216b354

- Replace `let _ = app.emit(...)` with logged warnings so dropped chat-done/chat-token events surface in logs instead of silently hanging the frontend
- Switch streaming buffer from String to Vec<u8> so multi-byte UTF-8 sequences split across TCP chunks are accumulated correctly before converting to str
- Use `.context("stream read error")` (anyhow) instead of manual anyhow::anyhow! for the chunk read error path
- Rename test parse_empty_token_is_handled -> parse_empty_content_chunk_deserializes (more precise)
- Add malformed_json_fails_to_parse test

feat(m2): add chat Tauri command — spawns Ollama stream, emits chat-token/done/error 9175b7a247

feat(m2): register llm module and chat command in lib.rs 0e3bfd24f1

feat(m2): wire ChatPanel to stream LLM responses — thinking state, token accumulation, error display cba4175260

pyr0ball added this to the M2: LLM Chat milestone 2026-05-19 07:35:45 -07:00

pyr0ball referenced this pull request

2026-05-19 07:46:44 -07:00

spec(M2): LLM Chat — Ollama-backed streaming chat with migration context #6

pyr0ball referenced this pull request

2026-05-19 10:45:37 -07:00

feat: full pattern matrix — M1 complete, M2 LLM chat, 30+ pattern files #10

This pull request can be merged automatically.

You are not authorized to merge this pull request.

View command line instructions

Checkout

From your project repository, check out a new branch and test the changes.

git fetch -u origin feat/m2-llm-chat:feat/m2-llm-chat

git checkout feat/m2-llm-chat

Merge

Merge the changes and update on Forgejo.

Warning: The "Autodetect manual merge" setting is not enabled for this repository, you will have to mark this pull request as manually merged afterwards.