Plugin endpoints

Each plugin's routes mount under /api/plugins/{plugin-name}/.

Assessment plugin

GET /api/plugins/assessment/questions?lang=en

Returns the 12-question pack with text resolved to the requested language. Each answer carries the per-method weights.

[
  {
    "id": "q01",
    "type": "multi",
    "text": "How do you approach a new topic?",
    "answers": [
      {
        "id": "a",
        "text": "I read the rules and theory first.",
        "weights": {"deductive": 1.0}
      },
      ...
    ]
  },
  ...
]

POST /api/plugins/assessment/evaluate

Body:

{
  "project_id": "p1",
  "answers": [
    {"question_id": "q01", "answer_ids": ["a", "b"]},
    {"question_id": "q02", "answer_id": "c"},
    ...
  ]
}

Both single-select (answer_id: string) and multi-select (answer_ids: string[]) shapes are accepted. Returns the created LearningProfile:

{
  "id": "pr1",
  "user_id": "abc-123",
  "project_id": "p1",
  "deductive": 0.4167,
  "inductive": 0.1833,
  "error_based": 0.0833,
  "dialogic": 0.0833,
  "contextual": 0.0833,
  "ai_adaptive": 0.1500,
  "assessed_at": "2026-05-19T12:00:00+00:00",
  "version": 1,
  "dominant_method": "deductive"
}

GET /api/plugins/assessment/profile/{project_id}

Returns the latest LearningProfile for the project.

Session plugin

POST /api/plugins/session/start

Body:

{
  "project_id": "p1",
  "method": "deductive",
  "cycle_step": 1,
  "lang": "en"
}

method, cycle_step, lang are optional. Returns the created session + the composed system prompt:

{
  "session": {
    "id": "s1",
    "project_id": "p1",
    "method": "deductive",
    "started_at": "2026-05-19T12:00:00+00:00",
    "ended_at": null,
    "cycle_step": 1,
    "status": "active"
  },
  "system_prompt": "You are a deductive learning companion. ..."
}

POST /api/plugins/session/{session_id}/message

Body:

{"role": "user", "content": "I'd like to start with the present tense."}

The server saves the user message, fires ai_complete, persists the assistant reply, runs the step evaluator, and returns the composite:

{
  "user_message": {"id": "m1", "session_id": "s1", "role": "user", "content": "...", "created_at": "..."},
  "assistant_message": {"id": "m2", "session_id": "s1", "role": "assistant", "content": "...", "created_at": "..."},
  "ai_error": null,
  "session": {"...": "session row after cycle_step advance"},
  "step_evaluation": {
    "advance": true,
    "confidence": 0.85,
    "reason": "Learner clearly grasped the input.",
    "suggested_step": 2,
    "fallback_used": false,
    "applied": true,
    "from_step": 1
  }
}

POST /api/plugins/session/{session_id}/rate

Body:

{
  "understanding": 4,
  "stress": 2,
  "method_fit": 5,
  "notes": "Felt productive."
}

Persists a SessionRating row. Returns the row.

POST /api/plugins/session/{session_id}/end

Closes the session. Fires on_session_complete (tracking plugin writes a ProgressCommit). Returns:

{"session": {"...": "the session with status='completed' and ended_at set"}}

GET /api/plugins/session/switch-recommendation/{session_id}

Returns:

{"recommended": false, "to_method": null, "reason": null}

Or, when stagnation is detected:

{"recommended": true, "to_method": "dialogic", "reason": "Three sessions of flat understanding + high stress."}

POST /api/plugins/session/{session_id}/switch

Body:

{"to_method": "dialogic", "reason": "Felt right."}

Updates session.method, writes a MethodSwitch audit row, returns the updated session.

Tracking plugin

GET /api/plugins/tracking/progress/{project_id}

Returns the namespaced summary; the tracking slice carries the aggregator's output:

{
  "tracking": {
    "total_sessions": 7,
    "total_minutes": 195,
    "streak_days": 3,
    "sessions_per_method": {"deductive": 4, "inductive": 2, "dialogic": 1},
    "method_distribution": [...],
    "recent_understanding": [0.6, 0.8, 0.8, 0.8, 1.0],
    "recent_stress": [0.4, 0.4, 0.2, 0.2, 0.2],
    "mean_understanding": 0.8,
    "mean_stress": 0.3,
    "recent_sessions": [...]
  },
  "step_evaluation": {
    "total_evaluations": 28,
    "average_confidence": 0.78,
    "advance_count": 21,
    "repeat_count": 5,
    "backward_count": 2,
    "fallback_count": 1,
    "evaluations_per_step": {"1": 7, "2": 6, ...},
    "time_seconds_per_step": {"1": 180.5, ...}
  }
}

GET /api/plugins/tracking/commits/{project_id}

Returns all ProgressCommit rows for the project, chronological.

Tools plugin

GET /api/plugins/tools/recommendations/{project_id}?lang=en

Returns the 5 catalog tools ranked by relevance to the project's profile:

[
  {
    "name": "Anki",
    "url": "https://apps.ankiweb.net/",
    "why": "Spaced-repetition flashcards - great for cementing rules and error-corrections long-term.",
    "weight_keys": ["deductive", "error_based"],
    "score": 0.5
  },
  ...
]

GET /api/plugins/tools/spaced/{project_id}?lang=en

Returns spaced-repetition action cards driven by recency:

[
  {
    "id": "sr-deductive-first",
    "method": "deductive",
    "interval_days": 1,
    "action": "session",
    "title": "First practice in deduction.",
    "urgency": 0.5
  },
  ...
]

Session plugin — streaming + pronunciation (v1.6.0+, v1.18.0+)

POST /api/plugins/session/{id}/message/stream  (SSE)

Same body shape as /message; emits three SSE event types:

• start — payload {user_message} (the user turn now persisted).
• chunk — payload {delta} (one or more text chunks arriving from the AI provider's stream).
• done — payload identical to the synchronous /message response: assistant message + cycle_step + timings + optional cycle-transition card.

GET  /api/plugins/session/pronunciation/eligibility/{project_id}
POST /api/plugins/session/pronunciation/phrase
POST /api/plugins/session/pronunciation/judge

Pronunciation eligibility is gated by the project's subject taxonomy (walks ancestors looking for a Languages / Sprachen root). The judge endpoint returns {matches, score, feedback, missed_sounds}.

Gamification plugin (v1.16.0+)

GET  /api/plugins/gamification/xp/{user_id}
POST /api/plugins/gamification/xp/{user_id}/award
POST /api/plugins/gamification/xp/{user_id}/award-assessment
POST /api/plugins/gamification/xp/{user_id}/award-import
GET  /api/plugins/gamification/badges
GET  /api/plugins/gamification/badges/{user_id}
POST /api/plugins/gamification/badges/{user_id}/evaluate
GET  /api/plugins/gamification/streak/{user_id}
GET  /api/plugins/gamification/streak/{user_id}/heatmap
POST /api/plugins/gamification/streak/{user_id}/weekend-mode
POST /api/plugins/gamification/reset/{user_id}

/streak/{user_id}/heatmap returns [{date, count}, ...] for the last 365 days (clamps to [7, 730]). The reset endpoint double-confirms then wipes user_xp + user_badges + user_streaks rows.

Anki plugin (v1.17.0+)

GET    /api/plugins/anki/cards/{user_id}
POST   /api/plugins/anki/cards
PATCH  /api/plugins/anki/cards/{id}
DELETE /api/plugins/anki/cards/{id}
POST   /api/plugins/anki/cards/extract/session/{id}
POST   /api/plugins/anki/cards/extract/conversation/{id}
POST   /api/plugins/anki/cards/mark-exported

AI extraction is user-triggered. The conversation-extract path also reads analysis_result.vocabulary (since v1.20.0) to produce cloze cards without an extra AI call when the analysis already populated the vocabulary array.

NotebookLM plugin (v1.19.0+)

GET    /api/plugins/notebooklm/questions/{user_id}
POST   /api/plugins/notebooklm/questions
PATCH  /api/plugins/notebooklm/questions/{id}
DELETE /api/plugins/notebooklm/questions/{id}
POST   /api/plugins/notebooklm/generate-from-session/{id}
POST   /api/plugins/notebooklm/generate-from-project/{id}
POST   /api/plugins/notebooklm/study-guide/{project_id}

/study-guide/{project_id} returns text/markdown (one big AI call with content-clipping to ~30K chars). User edits flip edited=True so the AI re-runner skips them.

Plugin discovery

GET /api/plugins/manifests
GET /api/plugins/health
GET /api/plugins/errors

Each returns a map keyed by plugin name. Used by Settings > Plugins UI for activation status + load-error visibility.