API (dev-api.py):
- GET /api/settings/fine-tune/pairs — list pairs from JSONL with index/instruction/source_file
- DELETE /api/settings/fine-tune/pairs/{index} — remove a pair and rewrite JSONL
- POST /api/settings/fine-tune/submit — now queues prepare_training task (replaces UUID stub)
- GET /api/settings/fine-tune/status — returns pairs_count from JSONL (not just DB task)
Store (fineTune.ts):
- TrainingPair interface
- pairs, pairsLoading refs
- loadPairs(), deletePair() actions
Vue (FineTuneView.vue):
- Step 2 shows scrollable pairs list with instruction + source file
- ✕ button on each pair calls deletePair(); list/count update immediately
- loadPairs() called on mount
API additions (dev-api.py):
- GET /api/tasks — list active background tasks
- DELETE /api/tasks/{task_id} — per-task cancel
- POST /api/tasks/kill — kill all stuck tasks
- POST /api/tasks/discovery|email-sync|enrich|score|sync — queue/trigger each workflow
- POST /api/jobs/archive — archive by statuses array
- POST /api/jobs/purge — hard delete by statuses or target (email/non_remote/rescrape)
- POST /api/jobs/add — queue URL imports
- POST /api/jobs/upload-csv — upload CSV with URL column
- GET /api/config/setup-banners — list undismissed onboarding hints
- POST /api/config/setup-banners/{key}/dismiss — dismiss a banner
HomeView.vue:
- 4th WorkflowButton: "Fill Missing Descriptions" (always visible, not gated on enrichment_enabled)
- Danger Zone redesign: scope radio (pending-only vs pending+approved), Archive & reset (primary)
vs Hard purge (secondary), inline confirm dialogs, active task list with per-task cancel,
Kill all stuck button, More Options (email purge / non-remote / wipe+rescrape)
- Setup banners: dismissible onboarding hints pulled from /api/config/setup-banners,
5-second polling for active task list to stay live
app/Home.py:
- Danger Zone redesign: same scope radio + archive/purge with confirm steps
- Background task list with per-task cancel and Kill all stuck button
- More options expander (email purge, non-remote, wipe+rescrape)
- Setup banners section at page bottom
API additions (dev-api.py split across this and next commit):
- /api/jobs/{job_id}/qa GET/PATCH/suggest — Interview Prep answer storage + LLM suggestions
- /api/settings/ui-preference POST — persist streamlit/vue preference to user.yaml
- cancel_task() added to scripts/db.py (per-task cancel for Danger Zone)
Vue / UI:
- AppNav: "⚡ Classic" button to switch back to Streamlit UI (writes cookie + persists to user.yaml)
- ApplyWorkspace: Resume Highlights panel (collapsible skills/domains/keywords with job-match highlighting)
- SettingsView: hide Data tab in cloud mode (showData guard)
- ResumeProfileView: minor improvements
- useApi.ts: error handling improvements
Infra:
- compose.cloud.yml: add api service (uvicorn dev_api running in cloud container)
- docker/web/nginx.conf: proxy /api/* to api service in cloud mode
- README.md: Vue SPA now listed as Free tier feature
- app/app.py: load_env at module level (safe in Docker, fills gaps on bare-metal)
- dev_api.py: load_env in startup handler (avoids test-env pollution)
- requirements.txt: note >= 0.7.0 requirement; TODO tag once cf-core cuts release
db.migration runner deferral: tracked in #43 (Rails-style numbered migrations)
CFOrchClient VRAM wiring: already present in task_scheduler via CF_ORCH_URL env var
When preflight.py adopts a host-running Ollama (or ollama_research) service,
write OLLAMA_HOST (and OLLAMA_RESEARCH_HOST) into .env using host.docker.internal
so LLMRouter's env-var auto-config resolves the correct address from inside the
Docker container without requiring a config/llm.yaml to exist.
#50 Job Review list view — sort + filter controls:
- Sort by best match / newest first / company A-Z (client-side computed)
- Remote-only checkbox filter
- Job count indicator; filters reset on tab switch
- Remote badge on list items
#61 Cover letter generation from approved tab:
- '✨ Draft' button on each approved-list item → /apply/:id
- No extra API call; ApplyWorkspace handles generation from there
#54 Company research modal (all API endpoints already existed):
- CompanyResearchModal.vue: 3-state machine (empty→generating→ready)
polling /research/task every 3s, displays all 7 research sections
(company, leadership, talking points, tech, funding, red flags,
accessibility), copy-to-clipboard for talking points, ↺ Refresh
- InterviewCard: new 'research' emit + '🔍 Research' button for
phone_screen/interviewing/offer stages
- InterviewsView: wires modal with researchJobId/Title/AutoGen state;
auto-opens modal with autoGenerate=true when a job is moved to
phone_screen (mirrors Streamlit behaviour)
- _user_yaml_path(): remove dangerous fallback to /devl/job-seeker/
config/user.yaml (Meg's legacy profile); a missing user.yaml now
returns an empty dict via load_user_profile, never another user's data
- RESUME_PATH: replace hardcoded relative Path('config/plain_text_
resume.yaml') with _resume_path() that derives from _user_yaml_path()
so resume file is always co-located with the correct user.yaml
- upload_resume: was passing a file path string to structure_resume()
which expects raw text; now extracts bytes, dispatches to the correct
extractor (pdf/odt/docx), then passes text — matches Streamlit wizard
- WizardResumeStep.vue: upload response is {ok, data: {experience…}}
but component was reading data.experience (top level); fixed to
read resp.data.experience to match the actual API envelope
New endpoints:
- GET /api/wizard/status — resume-after-refresh; returns wizard_step + saved_data
- POST /api/wizard/step — persist step data; side effects per step
(step 3: plain_text_resume.yaml, step 5: .env keys,
step 6: search_profiles.yaml)
- GET /api/wizard/hardware — GPU detection + profile suggestion
- POST /api/wizard/inference/test — soft-fail Ollama/LLM connectivity check
- POST /api/wizard/complete — set wizard_complete=true, apply service URLs
Updated:
- GET /api/config/app now includes wizardComplete (from user.yaml) and isDemo
(from DEMO_MODE env) so the Vue nav guard can gate on a single config fetch
30 tests, all passing
- Lower vue_ui_beta gate to "free" so all licensed users can access the
new UI without a paid subscription
- Remove "Paid tier" wording from the Try New UI banner
- Fix Vue SPA navigation in cloud/demo deployments: add VITE_BASE_PATH
build arg so Vite sets the correct subpath base, and pass
import.meta.env.BASE_URL to createWebHistory() so router links
emit /peregrine/... paths that Caddy can match
- Fix feedback button missing on cloud instance by passing
FORGEJO_API_TOKEN through compose.cloud.yml
- Remove vLLM container from compose.yml (vLLM dropped from stack;
cf-research service in cfcore covers the use case)
- Fix cloud config path in Apply page (use get_config_dir() so per-user
cloud data roots resolve correctly for user.yaml and resume YAML)
- Refactor generate_cover_letter._build_system_context and
_build_mission_notes to accept explicit profile arg (enables
per-user cover letter generation in cloud multi-tenant mode)
- Add API proxy block to nginx.conf (Vue web container can now call
/api/ directly without Vite dev proxy)
- Update .env.example: remove vLLM vars, add research model + tuning
vars for external vLLM deployments
- Update llm.yaml: switch vllm base_url to host.docker.internal
(vLLM now runs outside Docker stack)
Closes#63 (feedback button)
Related: #8 (Vue SPA), #50–#62 (parity milestone)
vllm is now managed by cf-orch as a host process — no Docker service
defined in compose.yml. Preflight was detecting port 8000 (llm_server)
and generating a vllm stub in compose.override.yml with no image,
causing `docker compose up` to error on startup.
Phi-4-mini's cached modeling_phi3.py imports SlidingWindowCache which
was removed in transformers 5.x. Qwen2.5-3B uses built-in qwen2 arch
and works cleanly. Reorder so Qwen is tried first.
Remove claude_code, github_copilot, and anthropic from all cloud fallback
orders — cloud accounts must not route through personal/dev LLM backends.
vllm_research and ollama_research are the only permitted research backends.
llm.cloud.yaml is now bind-mounted at /app/config/llm.yaml in compose.cloud.yml.
- discover.py: run_discovery() accepts config_dir param; auto-derives it
from db_path parent (per-user in cloud, falls back to /app/config)
- task_runner.py: passes db_path.parent/config as config_dir to run_discovery
- wizard (0_Setup.py): write 'titles' key not 'job_titles' — matches what
discover.py and all custom board scrapers read
- adzuna/theladders/craigslist: fall back to 'job_titles' for existing
profiles written by older wizard versions
- Fixed Sheridan's live config in place (job_titles → titles)
Module-level DB_PATH is frozen at import time, so monkeypatch.setenv()
in tests had no effect on _user_yaml_path(). Reading os.environ directly
fixes 6 test_dev_api_settings PUT/POST failures in CI where
/devl/job-seeker/ doesn't exist.
FORGEJO_TOKEN secret injected via env var (not inline expression) to
avoid CI injection risk; git insteadOf redirect authenticates the
git+https:// circuitforge-core VCS URL at install time.
CI checks out only the peregrine repo — ../circuitforge-core doesn't exist,
causing 'pip install -r requirements.txt' to fail.
requirements.txt now uses git+https://...@main as the fallback for CI and
bare pip installs. Dockerfile.cfcore installs cfcore from the local COPY
first (skipping the requirements.txt line) to avoid a redundant network
fetch during Docker builds.
compose.yml was still using build: . which fails because requirements.txt
has -e ../circuitforge-core outside the build context. Now matches
compose.cloud.yml and compose.test-cfcore.yml.
Switch compose.cloud.yml build context to Dockerfile.cfcore (parent
context includes circuitforge-core/ as sibling). Adds CF_ORCH_URL so
the cloud container can reach the cf-orch coordinator on the host.
Threads coordinator_url from CF_ORCH_URL env var (default localhost:7700)
into the cfcore TaskScheduler so Docker instances can point at
host.docker.internal:7700 instead of the container's own loopback.
Also adds CF_ORCH_URL to compose.test-cfcore.yml and mounts persistent
patched configs (llm.docker.yaml, user.docker.yaml) for the test instance.
sync_ui_cookie() was calling window.parent.location.reload() on every
render when user.yaml has ui_preference=vue, but no Caddy is in the
traffic path (test instances, bare Docker). This caused an infinite
reload loop because the reload just came back to Streamlit.
Gate the reload on PEREGRINE_CADDY_PROXY=1. Without it, the cookie is
still written silently but no reload is attempted. Add the env var to
compose.yml and compose.cloud.yml (both are behind Caddy); omit from
compose.test-cfcore.yml so test instances stay stable.
- Dockerfile: restored to original (build: . context, no cfcore) so
existing compose.yml / compose.cloud.yml builds are unaffected
- Dockerfile.cfcore: parent-context build that copies circuitforge-core/
alongside peregrine/ before pip install; resolves -e ../circuitforge-core
- compose.test-cfcore.yml: single-user test instance on port 8516;
run from parent dir with context: .. so both repos are in scope
Use this to smoke-test cfcore shims before promoting to prod cloud.
- Catch all exceptions (not just RuntimeError) so FileNotFoundError,
connection errors, etc. surface as error messages rather than crashing
the page silently
- Show "No new suggestions found" info message when the LLM returns
empty arrays — previously the spinner completed with no UI feedback
- Hint to upload resume when RESUME_PATH is missing (new users)
- Only rerun() when there are actual results to display
Pages were hardcoding DEFAULT_DB at import time, meaning cloud-mode
per-user DB routing was silently ignored. Pages affected:
1_Job_Review, 5_Interviews, 6_Interview_Prep, 7_Survey.
Adds resolve_session("peregrine") + get_db_path() pattern to each,
matching the pattern already used in 4_Apply.py.
Fixes#24.
importlib.reload(dev_api) reset all module-level globals (RESUME_PATH,
SEARCH_PREFS_PATH, etc.) on every digest/interviews test, causing
subsequent monkeypatch.setattr calls in test_dev_api_settings.py to
silently fail — the patched attribute was reset between fixture setup
and the actual HTTP request.
Fix: patch dev_api.DB_PATH directly via monkeypatch, which pytest reverts
cleanly after each test without touching any other module state.
Also sync resume optimizer endpoints to dev-api.py (hyphen variant).
ApplyWorkspace.vue: kept HEAD (vue-spa) version for resume optimizer panel,
cl-error__actions wrapper, and ResumeOptimizerPanel import. main's older
version lacked these additions.
