- preflight.py now writes PEREGRINE_GPU_COUNT and PEREGRINE_GPU_NAMES to
.env so the app container gets GPU info without needing nvidia-smi access
- compose.yml passes PEREGRINE_GPU_COUNT, PEREGRINE_GPU_NAMES, and
RECOMMENDED_PROFILE as env vars to the app service
- 0_Setup.py _detect_gpus() reads PEREGRINE_GPU_NAMES env var first;
falls back to nvidia-smi (bare / GPU-passthrough environments)
- 0_Setup.py _suggest_profile() reads RECOMMENDED_PROFILE env var first
- requirements.txt: add pdfplumber (needed for resume PDF parsing)
Without __init__.py, Python treats app/ as a namespace package that
doesn't resolve correctly when running from WORKDIR /app inside the
container. 'from app.wizard.step_hardware import ...' raises
ModuleNotFoundError: No module named 'app.wizard'; 'app' is not a package
Three inter-related fixes for the service adoption flow:
- preflight: stub_port field — adopted services get a free port for their
no-op container (avoids binding conflict with external service on real port)
while update_llm_yaml still uses the real external port for host.docker.internal URLs
- preflight: write_env now uses stub_port (not resolved) for adopted services
so SEARXNG_PORT etc point to the stub's harmless port, not the occupied one
- preflight: stub containers use sleep infinity + CMD true healthcheck so
depends_on: service_healthy is satisfied without holding any real port
- Makefile: finetune profile changed from [cpu,single-gpu,dual-gpu] to [finetune]
so the pytorch/cuda base image is not built during make start
- preflight: ollama was incorrectly marked docker_owned=False — Docker does
define an ollama service, so external detection now correctly disables it
via compose.override.yml when host Ollama is already running
- compose.yml: finetune moves from [cpu,single-gpu,dual-gpu] profiles to
[finetune] profile so it is never built during 'make start' (pytorch/cuda
base is 3.7GB+ and unnecessary for the UI)
- compose.yml: remove depends_on ollama from finetune — it reaches Ollama
via OLLAMA_URL env var which works whether Ollama is Docker or host
- Makefile: finetune target uses --profile finetune + compose.gpu.yml overlay
preflight.py now detects when a managed service (ollama, vllm, vision,
searxng) is already running on its configured port and adopts it rather
than reassigning or conflicting:
- Generates compose.override.yml disabling Docker containers for adopted
services (profiles: [_external_] — a profile never passed via --profile)
- Rewrites config/llm.yaml base_url entries to host.docker.internal:<port>
so the app container can reach host-side services through Docker's
host-gateway mapping
- compose.yml: adds extra_hosts host.docker.internal:host-gateway to the
app service (required on Linux; no-op on macOS Docker Desktop)
- .gitignore: excludes compose.override.yml (auto-generated, host-specific)
Only streamlit is non-adoptable and continues to reassign on conflict.
- Remove ollama-gpu service (was colliding with ollama on port 11434)
- Strip inline deploy.resources GPU blocks from vision and vllm
- Add compose.gpu.yml: Docker NVIDIA overlay for ollama (GPU 0),
vision (GPU 0), vllm (GPU 1), finetune (GPU 0)
- Fix compose.podman-gpu.yml: rename ollama-gpu → ollama to match
service name after removal of ollama-gpu
- Update Makefile: apply compose.gpu.yml for Docker + GPU profiles
(was only applying podman-gpu.yml for Podman + GPU profiles)
- setup.sh: replace docker-image-based NVIDIA test with nvidia-ctk validate
(faster, no 100MB pull, no daemon required); add check_docker_running()
to auto-start the Docker service on Linux or warn on macOS
- prepare_training_data.py: also scan training_data/uploads/*.{md,txt}
so web-uploaded letters are included in training data
- task_runner.py: add prepare_training task type (calls build_records +
write_jsonl inline; reports pair count in task result)
- Settings fine-tune tab: Step 1 accepts .md/.txt uploads; Step 2 Extract
button submits prepare_training background task + shows status; Step 3
shows make finetune command + live Ollama model status poller
- Dockerfile.finetune: PyTorch 2.3/CUDA 12.1 base + unsloth + training stack
- finetune_local.py: auto-register model via Ollama HTTP API after GGUF
export; path-translate between finetune container mount and Ollama's view;
update config/llm.yaml automatically; DOCS_DIR env override for Docker
- prepare_training_data.py: DOCS_DIR env override so make prepare-training
works correctly inside the app container
- compose.yml: add finetune service (cpu/single-gpu/dual-gpu profiles);
DOCS_DIR=/docs injected into app + finetune containers
- compose.podman-gpu.yml: CDI device override for finetune service
- Makefile: make prepare-training + make finetune targets
Interactive prompt lets users with split-drive setups point Ollama and
vLLM model dirs at a dedicated storage drive. Reads current .env value
as default so re-runs are idempotent. Skips prompts in non-interactive
(piped) mode. Creates the target directory immediately and updates .env
in-place via portable awk (Linux + macOS). Also simplifies next-steps
output since model paths are now configured at install time.
- llm.yaml + example: replace localhost URLs with Docker service names
(ollama:11434, vllm:8000, vision:8002); replace personal model names
(alex-cover-writer, llama3.1:8b) with llama3.2:3b
- user.yaml.example: update service hosts to Docker names (ollama, vllm,
searxng) and searxng port from 8888 (host-mapped) to 8080 (internal)
- wizard step 5: fix hardcoded localhost defaults — wizard runs inside
Docker, so service name defaults are required for connection tests to pass
- scrapers/companyScraper.py: bundle scraper so Dockerfile COPY succeeds
- setup.sh: remove host Ollama install (conflicts with Docker Ollama on
port 11434); Docker entrypoint handles model download automatically
- README + setup.sh banner: add Circuit Forge mission statement
- generate() accepts previous_result + feedback; appends both to LLM prompt
- task_runner cover_letter handler parses params JSON, passes fields through
- Apply Workspace: "Refine with Feedback" expander with text area + Regenerate
button; only shown when a draft exists; clears feedback after submitting
- 8 new tests (TestGenerateRefinement + TestTaskRunnerCoverLetterParams)
Replaces the old 5-step wizard with a 7-step orchestrator that uses the
step modules built in Tasks 2-8. Steps 1-6 are mandatory (hardware, tier,
identity, resume, inference, search); step 7 (integrations) is optional.
Each Next click validates, writes wizard_step to user.yaml for crash recovery,
and resumes at the correct step on page reload. LLM generation buttons
submit wizard_generate tasks and poll via @st.fragment(run_every=3). Finish
sets wizard_complete=True, removes wizard_step, and calls apply_service_urls.
Adds tests/test_wizard_flow.py (7 tests) covering validate() chain, yaml
persistence helpers, and wizard state inference.
Add all 13 integration modules (Notion, Google Drive, Google Sheets,
Airtable, Dropbox, OneDrive, MEGA, Nextcloud, Google Calendar, Apple
Calendar/CalDAV, Slack, Discord, Home Assistant) with fields(), connect(),
and test() implementations. Add config/integrations/*.yaml.example files
and gitignore rules for live config files. Add 5 new registry/schema
tests bringing total to 193 passing.
- Add tier, dev_tier_override, wizard_complete, wizard_step, dismissed_banners
fields to UserProfile with defaults and effective_tier property
- Add params TEXT column to background_tasks table (CREATE + migration)
- Update insert_task() to accept params with params-aware dedup logic
- Update submit_task() and _run_task() to thread params through
- Add test_wizard_defaults, test_effective_tier_override,
test_effective_tier_no_override, and test_insert_task_with_params
scripts/preflight.py (stdlib-only, no psutil):
- Port probing: owned services auto-reassign to next free port; external
services (Ollama) show ✓ reachable / ⚠ not responding
- System resources: CPU cores, RAM (total + available), GPU VRAM via
nvidia-smi; works on Linux + macOS
- Profile recommendation: remote / cpu / single-gpu / dual-gpu
- vLLM KV cache offload: calculates CPU_OFFLOAD_GB when VRAM < 10 GB
free and RAM headroom > 4 GB (uses up to 25% of available headroom)
- Writes resolved values to .env for docker compose; single-service mode
(--service streamlit) for scripted port queries
- Exit 0 unless an owned port genuinely can't be resolved
scripts/manage-ui.sh:
- Calls preflight.py --service streamlit before bind; falls back to
pure-bash port scan if Python/yaml unavailable
compose.yml:
- vllm command: adds --cpu-offload-gb ${CPU_OFFLOAD_GB:-0}
Makefile:
- start / restart depend on preflight target
- PYTHON variable for env portability
- test target uses PYTHON variable
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
