What can Midcore do?

1. Outcome Compiler · Vibe Coding

AI coding IDE — read, write, refactor, debug, plan, and ship code against the user's workspace.

The interactive coding surface. The agent has read/write access to your open workspace via file + shell + git + search tools; runs the full Outcome Compiler pipeline (intent → scope → execute → evidence) on demand; and surfaces every change with diffs, lint results, and gate signals.

What I'll drive for you: build, debug, refactor, and ship code. Tell me what outcome you want; I'll read the existing code, draft a plan, edit the files, run the tests, and report what changed. If the change requires explicit approval (force-push, external deploy, destructive shell), I'll pause and ask first.

Example tasks I'll walk you through:

• Build a full app from scratch (build_full_app)
• Debug a failing test (debug_failing_test)
• Refactor a module (refactor_module)
• Add a feature to existing code (add_feature)
• Migrate between frameworks (migrate_framework)

Routes: /api/v1/autonomy/vibe/{run|stream|cancel|rollback|tool-feedback} · Docs: /docs/concepts/outcome-compiler

2. Automations

Durable multi-agent automations that wake on schedule, webhook, or inbound event.

Postgres-backed automation platform. 10-table schema, scheduler + worker daemon with SKIP LOCKED claims, LISTEN/NOTIFY inter-agent messaging, 8 customer-ready templates, 11 SaaS integration adapters (Slack, Notion, HubSpot, Linear, X/Twitter, LinkedIn, Gmail/Outlook OAuth, generic SMTP/ IMAP, Google Calendar, HTTP, RSS), and HMAC-verified webhook intake for GitHub / Slack / Linear / PagerDuty / email / generic.

What I'll drive for you: create, configure, enable, run, and monitor automations. Pick one of the 8 customer-ready templates or describe what you want and I'll wire the triggers, attach the credentials, and run it durably — your laptop can close and the work keeps moving.

Example tasks I'll walk you through:

• Set up a daily morning brief (setup_daily_brief)
• Wire up a Slack-triggered automation (wire_slack_automation)
• Connect a new SaaS integration (connect_saas_integration)
• Migrate legacy filesystem automations (migrate_fs_to_db)
• Set up the Customer Support Triager (setup_support_triager)

Routes: /api/v1/automations/*, /api/v1/sessions/*, /webhooks/{tenant_id}/{provider} · Docs: /docs/guides/automation-setup, /docs/reference/integrations

3. Research

Scout · Simulation · Survey · Pharmacovigilance Surveillance across 20+ domains — every track finalizable with auditable real outputs, first-world regulatory tooling, and per-jurisdiction privacy gates.

The research workspace covers four product tracks plus 20+ domain templates (medicine, pharmacovigilance, genetics, physics signals, telecom, climate, finance, neuroscience, ecology, agriculture, energy, materials, robotics control, aerospace, manufacturing, OR, social science, education, cybersecurity, ML, CUSTOM). Every track ends with a real Complete session action that writes a stage="completed" completion record — refused unless real outputs exist (412 if Monte Carlo never ran, no real hypotheses, no synthesis, or no signals detected). Simulation uses a real local Monte Carlo executor (deterministic seed, convergence-aware) — never stubs.

Pharmacovigilance Surveillance is regulator-grade and appealing to first-world PV teams (FDA, EMA, Health Canada, TGA, MHRA, PMDA): WHO-UMC / Naranjo / RUCAM causality scoring, proper Bayesian BCPNN IC025 (Bate 1998 — the EudraVigilance / WHO-VigiBase convention), subgroup stratification (gender / age / region), E2B(R3) ICSR XML import + export with safe parser, CIOMS I print-ready forms, PSUR/PBRER aggregate-report assembly per ICH E2C(R2), and jurisdiction-aware privacy rules (HIPAA / PIPEDA / GDPR / Privacy Act 1988 / UK GDPR / Japan APPI) with cross-border transfer gates.

What I'll drive for you: a complete research session end-to-end. Pick a track: SIMULATION (wizard → brief → evidence → Monte Carlo → analysis → completion record), SCOUT (anti-fabrication hypothesis exploration), SURVEY (PRISMA + GRADE systematic review), or PHARMACOVIGILANCE SURVEILLANCE (protocol → FAERS/E2B ingest → screening → BCPNN IC025 signal detection → subgroup stratification → WHO-UMC/Naranjo/ RUCAM causality scoring → assessment → CIOMS I / E2B(R3) / PSUR output, all with HIPAA/GDPR/PIPEDA-aware redaction and cross-border gates).

Example tasks I'll walk you through:

• Explore a research domain from scratch (research_explore_domain)
• Run a Monte Carlo simulation (research_monte_carlo)
• Finalize a research session end-to-end (research_session_finalize)
• Compile a public-source dossier on a person/organization (research_dossier_grounded)
• Open a pharmacovigilance surveillance session (research_surveillance)
• Score causality for an adverse drug reaction (pv_causality_assessment)
• Compute disproportionality signals (PRR / ROR / IC025) (pv_signal_detection)
• Stratify a drug-event signal by gender / age / region (pv_subgroup_analysis)
• Import or export E2B(R3) ICSR XML (pv_e2b_codec)
• Generate a CIOMS I form, print-ready (pv_cioms_i)
• Assemble a PSUR / PBRER periodic safety report (pv_psur)
• Check cross-border transfer eligibility for a case (pv_jurisdiction_gate)
• Compare LLM reasoning across models (research_debate)

Routes: /api/v1/autonomy/research/{wizard|scout|survey|surveillance}/..., /api/v1/autonomy/research/sessions/{id}/finalize, /api/v1/autonomy/research/surveillance/{id}/{causality/score|subgroups/analyze|cioms-i|psur/generate}, /api/v1/autonomy/research/surveillance/{e2b/import|e2b/export|jurisdictions} · Docs: /docs/product/research, /docs/reference/grounded-research

4. Robotics · Operational Puble

Fleet, missions, planning (A*/RRT), safety gates, telemetry, and training from runs.

The robotics tier proxies through the Rust puble-server for real-time motion + safety; the Python layer owns fleet coordination, mission planning, training (baseline snapshots + heuristic learning from runs), and integrations (ROS, MQTT, MAVLink). Mission planners include waypoint, A* grid, and RRT; safety gates include estop and validate-before-execute.

What I'll drive for you: plan missions, coordinate a fleet, validate safety envelopes, ingest telemetry, and train a robot from past runs. Tell me the platform (drone / arm / mobile base / custom), the scenario, and the constraints — I'll lay out a plan and drive the runs. Destructive actions (deploy to hardware, override safety gate) always pause for explicit approval.

Example tasks I'll walk you through:

• Train a robot on a unique scenario (robotics_train_scenario)
• Plan a multi-waypoint mission (robotics_plan_waypoint_mission)
• Coordinate a fleet across a goal (robotics_fleet_coordinate)
• Validate safety envelope before deployment (robotics_safety_validate)
• Run a simulation before deploying (robotics_simulate_first)

Routes: /api/v1/autonomy/robotics/* (fleet, mission, safety, telemetry, training, simulation) · Docs: /docs/product/robotics

How the chat agent uses this

Three invariants the agent guarantees on every surface (web, desktop, CLI, VS Code):

1. Knows the four tiers. The capability manifest is injected into every system prompt header so the agent never "forgets" that it can do research or robotics — even mid-coding-session.
2. Answers "how do I X" with a concrete plan. The agent calls midcore_walkthrough for any "how do I / walk me through / step by step" prompt and surfaces the structured plan (prerequisites, numbered steps with how + why, gotchas, verification).
3. Acts on your behalf when asked. If you say "do X for me," the agent drives the matching tools (file/shell/git for coding, automation_create_from_template for automations, research_start_session for research, robotics_plan_mission for robotics). Destructive actions always pause for explicit approval.