Roadmap
What's shipped, what's in progress, and what's planned.
Shipped
| Feature | What landed |
|---|---|
| Web Frontend | Next.js chat UI with streaming, persisted setup state, thread management, memory inspection, visible error fallbacks, and OpenAI Realtime voice entrypoint. Lives in apps/web/. |
| API Layer | FastAPI with REST (POST /api/chat) and WebSocket (/api/chat/stream) endpoints. Thread management, memory status, session end. Lives in apps/backend/api/. |
| Voice Chat (OpenAI Realtime) | Browser speech-to-speech over OpenAI Realtime WebRTC with app-owned tools, Realtime session policy, incognito/persistent modes, turn recording, and shared end-session finalization. Lives across apps/web/src/components/realtime-voice-session-provider.tsx and apps/backend/agent/voice/. |
| Session Feedback | End-of-session thumbs rating captured at /end, /exit, and POST /api/threads/{id}/end. PostgresSessionFeedbackBackend is the durable implementation; incognito mode uses InMemorySessionFeedbackBackend. The former SQLite feedback implementation has been removed. |
| Crisis Gate — LLM-only | Crisis classification is a structured LLM call with strict truth-table enforcement. Provider failures surface through retries/errors instead of silently degrading to regex rules. |
| Routing — LLM-primary | Crisis, therapeutic dispatch, grounded lookup, memory-control, guided-exercise selection, and memory write policy use LLM-owned classifiers with local validation and hard confirmation gates where needed. |
| Knowledge Overhaul | core_identity.md defines assistant role, product stance, voice, therapeutic grounding, cultural sensitivity, repair patterns, and boundary-setting voice. boundaries.md expands redirection patterns and dependency framing. |
| OpenAI Embeddings | text-embedding-3-large as the configured provider, with token-only retrieval when no API key is available. Hybrid RRF retrieval achieves 14/17 recall@5 vs 6/17 token-only. |
In progress
| Feature | Status | What's left |
|---|---|---|
| Response quality review | Manual dogfood path | Needs broader review of ordinary support turns and longer dogfood transcripts. |
| Memory integration regression coverage | Pytest-first | Runtime and recall tests cover semantic, episodic, procedural, correction, deletion, and cross-feature behavior. Remaining work is wider live dogfood coverage and voice parity. |
| Session feedback — closing mode | Closing signal wired, feedback UX pending | Closing detection is LLM-primary and emits session_action=suggest_end_session; feedback prompt still needs to fire from natural closings, not just CLI/API end commands. |
| Session feedback — voice | Designed, not wired | Voice disconnect now routes through end_session() in persistent mode; the remaining work is collecting an explicit feedback value from the voice end-state UI. |
Planned
Messaging Channels
WhatsApp, Discord, and Telegram adapters. The text runtime contract is
channel-agnostic; each adapter would map platform message formats
to AgentInput / AgentOutput. Crisis responses would need
channel-specific formatting (inline buttons, embeds).
Acoustic Crisis Detection
Voice mode currently uses transcript-only crisis detection. Real gaps: voice cracking, sobbing, pressured speech, prosodic flatness. A user saying "I'm fine" through tears scores level 0.
Requires either a curated distressed-voice dataset (ethically fraught) or a validated off-the-shelf acoustic classifier (not a solved problem). Calendar-gated on dataset and model maturity.
Graph Memory
Graphiti + Neo4j for entity/relationship extraction from semantic
facts. Enables relational reasoning: "you mentioned your sister and
your work stress — they tend to co-occur." The graphiti-core
dependency is in pyproject.toml, but the integration is
intentionally disabled pending design.
Background Consolidation
Automatic fact merging, dormant marking, and a consolidation_runs
log. Schema is defined (ConsolidationProposal,
ConsolidationRunRecord in agent/memory/models.py); the
implementation is planned but not wired into the runtime. Adds
/memory restore as an undo for destructive operations.
Session Intent, Stage, and Response Guidance
Three state fields (progress.intent, progress.stage,
response.guidance) are defined in the schema but not yet populated
by the runtime. When implemented, they enable session-level steering:
the agent knows whether to deepen, stabilize, or close based on
conversation arc rather than just the current message.
Crisis Gate Production Telemetry
Model ID, prompt version, raw/normalized levels, confidence values, timeout/parse failure counters, and degraded-mode alerts. The production telemetry layer is not yet in place.
Clinical Review
A trained clinician reviews the agent/prompts/sources/response_styles/*.md
files, the agent-owned prompt builders in agent/specialists/, and
agent responses across dogfood sessions. This is the gate before
"a trusted friend could try it" becomes a defensible claim. Calendar
dependency, not engineering.