Auto-start CPU sandboxes for sessions (#200)

Add session YOLO auto-approval budget (#201)

Steer agent to HF kernels instead of pip install flash-attn (#204)

Use HF username for personal trace uploads (#199)

Auto-share session traces to a private HF dataset (#193)

Add DeepSeek V4 Pro model option (#189)

Make sandboxes private (#195)

feat(telemetry): track 5 untracked Bedrock call sites for full cost attribution (#179)

Track Pro conversions + credits top-up; emit user_id + new KPIs (#174)

Bill HF Jobs by namespace credits, drop Pro and org-join gating (#172)

Persist Space sessions in MongoDB (#170)

Update system prompt to reflect new role title (#169)

fix(sandbox): retry cleanup on transient failures + add backstop sweeper (#163)

Recognize Pro OAuth users via isPro flag (#164)

Embed trackio dashboard in chat for jobs and sandboxes (#161)

Add web search so the agent can cite current sources (#159)

Document Trackio alerts + iteration loop in system prompt (#156)

Add Slack gateway (#116)

fix(session-uploader): forward user_id and total_cost_usd to dataset row (#153)

Honor interactive model override after current CLI changes (#111)

Use `huggingface_hub.get_token()` for auth fallback (#126)

Protect sandbox control-plane routes (#142)

Preserve streamed thinking metadata with live model tests (#150)

Preserve thinking state across tool turns (#143)

Keep repetition guard internal (#144)

fix(doom_loop): normalize tool-call args before hashing (#119)

Treat Trackio as core for training and prefer public Space dashboards (#129)

fix(cli): persist HF user id in saved sessions (#146)

feat(session): include user_id and total_cost_usd in trajectory dump (#136)

Run jobs under user or org accounts with upgrade / org-pick UX (#132)

Make the CLI startup banner show the actual CLI default model (#122)

Fix CLI rendering corruption and split CLI/frontend model defaults (#121)

Switch from Bedrock to Anthropic endpoint as default. Include support for gpt-5.5 (#118)

feat(observability)

fix: route bedrock/ model ids via LiteLLM Bedrock adapter (#89)

feat: route Claude via Bedrock inference profile (#82)

feat: enable Anthropic prompt caching on system prompt and tools (#69)

fix /model switcher with effort params

Make X-HF-Bill-To configurable via HF_BILL_TO env (default smolagents) (#57)

Restore expired sessions via summary-based catch-up (#52)

Read max context from litellm + compact at 90% (#54)

Fix parallel-research display: per-call stats in CLI and web UI (#49)

Merge pull request #50 from huggingface/fix/ctrl-c-and-streaming-rework

Merge pull request #51 from huggingface/feat/hf-auto-routing

Fix CLI particle logo text: ML AGENT → ML INTERN (#46)

Strip provider_specific_fields from research sub-agent messages

Route HF inference through /v1 auto-router + add reasoning_effort knob

CLI: global Ctrl+C handling, progressive streaming, headless cleanup

CLI /model: accept any inference-provider model id

Fall back to user HF token for router when INFERENCE_TOKEN is unset (#39)