Frida is a working AI runtime focused on dialogue, memory, hermeneutic judgment, and operator observability. This repository tracks the real engineering state of the system as of April 3, 2026.
Primary references for the current repository state:
app/docs/states/project/Frida-State-english-03-04-26.mdapp/docs/states/project/Frida-State-french-03-04-26.mdapp/docs/todo-done/audits/fridadev_repo_audit.mdapp/docs/README.md
- Flask runtime exposing
/,/admin,/log,/hermeneutic-admin,/api/chat,/api/admin/settings/*,/api/admin/hermeneutics/*, and the live dashboard routeGET /api/admin/hermeneutics/dashboard. - Verified local stack on
http://127.0.0.1:8093, currently running withHERMENEUTIC_MODE=enforced_all. - Chat flow wired as: session resolution -> time grounding -> memory retrieval/arbitration ->
stimmung_agent->primary_node->validation_agent->[JUGEMENT HERMENEUTIQUE]injection -> main LLM call -> persistence and logs. - Operator observability keeping local
estimated_*counters distinct from post-call OpenRouterprovider_*truth. - OpenRouter transport identity split by caller (
llm,arbiter,identity_extractor,resumer,stimmung_agent,validation_agent) through dedicatedHTTP-RefererandX-OpenRouter-Title. - Main system prompt now framing Frida as a work/reflection interlocutor rather than a generic execution assistant.
- Main chat surface
/remains plain-text; the main LLM is instructed to answer in strict plain text, without visible Markdown formatting unless code is explicitly requested.
- Flask backend orchestration in
app/server.py - Core chat flows in
app/core/ - Memory and identity pipeline in
app/memory/andapp/identity/ - Admin/runtime settings and hermeneutic services in
app/admin/ - Observability modules in
app/observability/ - Web UI in
app/web/ - Tests in
app/tests/ - Structured documentation in
app/docs/
- Current EN project state:
app/docs/states/project/Frida-State-english-03-04-26.md - Current FR project state:
app/docs/states/project/Frida-State-french-03-04-26.md - Canonical repo audit:
app/docs/todo-done/audits/fridadev_repo_audit.md - Installation/operations guide:
app/docs/states/operations/frida-installation-operations.md - Active memory roadmap:
app/docs/todo-todo/memory/hermeneutical-add-todo.md - Active installation roadmap:
app/docs/todo-todo/product/Frida-installation-config.md - Closed Lot 9 archive:
app/docs/todo-done/refactors/hermeneutic-convergence-node-todo.md
Versioned:
- code, prompts, scripts, tests, docs
Not versioned:
app/.env- runtime state under
state/conv,state/logs,state/data - local caches, virtualenvs, OS/editor residue
Container mapping note:
/app/conv,/app/logs,/app/dataare container mount targets for those hoststate/...directories.
A fresh clone still needs a valid local .env, reachable runtime dependencies, and initialized local runtime state under state/....
./stack.sh up
./stack.sh ps
./stack.sh healthdocker-compose.ymlstack.shapp/server.pyapp/minimal_validation.pyapp/docs/README.md
- website: https://frida-ai.fr
- contact: tofmuck@frida-ai.fr
This repository is distributed under the MIT License. See LICENSE.
Frida est un runtime IA de travail centre sur le dialogue, la memoire, le jugement hermeneutique et l'observabilite operateur. Ce depot suit l'etat d'ingenierie reel du systeme au 3 avril 2026.
References principales pour l'etat actuel du depot:
app/docs/states/project/Frida-State-french-03-04-26.mdapp/docs/states/project/Frida-State-english-03-04-26.mdapp/docs/todo-done/audits/fridadev_repo_audit.mdapp/docs/README.md
- Runtime Flask exposant
/,/admin,/log,/hermeneutic-admin,/api/chat,/api/admin/settings/*,/api/admin/hermeneutics/*et la route liveGET /api/admin/hermeneutics/dashboard. - Stack locale verifiee sur
http://127.0.0.1:8093, actuellement enHERMENEUTIC_MODE=enforced_all. - Pipeline chat branche ainsi: resolution de session -> grounding temporel -> retrieval/arbitrage memoire ->
stimmung_agent->primary_node->validation_agent-> injection[JUGEMENT HERMENEUTIQUE]-> appel LLM principal -> persistance et logs. - Observabilite operateur distinguant les compteurs locaux
estimated_*et la verite OpenRouter post-callprovider_*. - Identite transport OpenRouter differenciee par composant (
llm,arbiter,identity_extractor,resumer,stimmung_agent,validation_agent) via desHTTP-RefereretX-OpenRouter-Titlededies. - Prompt systeme principal recadre comme interlocuteur de travail et de reflexion, et non comme simple assistant d'execution.
- La surface chat principale
/reste en texte brut ; le LLM principal est desormais contraint a repondre en texte brut strict, sans Markdown visible sauf si du code est explicitement demande.
- Orchestration backend Flask dans
app/server.py - Flux chat centraux dans
app/core/ - Pipeline memoire et identite dans
app/memory/etapp/identity/ - Services admin/runtime settings et hermeneutiques dans
app/admin/ - Modules d'observabilite dans
app/observability/ - UI web dans
app/web/ - Tests dans
app/tests/ - Documentation structuree dans
app/docs/
- Etat projet FR courant:
app/docs/states/project/Frida-State-french-03-04-26.md - Etat projet EN courant:
app/docs/states/project/Frida-State-english-03-04-26.md - Audit canonique du repo:
app/docs/todo-done/audits/fridadev_repo_audit.md - Guide installation/exploitation:
app/docs/states/operations/frida-installation-operations.md - Roadmap memoire active:
app/docs/todo-todo/memory/hermeneutical-add-todo.md - Roadmap installation active:
app/docs/todo-todo/product/Frida-installation-config.md - Archive de cloture Lot 9:
app/docs/todo-done/refactors/hermeneutic-convergence-node-todo.md
Versionne:
- code, prompts, scripts, tests, docs
Non versionne:
app/.env- etat runtime sous
state/conv,state/logs,state/data - caches locaux, virtualenvs et residus systeme/editeur
Repere conteneur:
/app/conv,/app/logs,/app/datasont les cibles de montage conteneur des repertoires hotestate/....
Un clone neuf necessite toujours un .env local valide, des dependances runtime joignables et un state local initialise sous state/....
./stack.sh up
./stack.sh ps
./stack.sh healthdocker-compose.ymlstack.shapp/server.pyapp/minimal_validation.pyapp/docs/README.md
- site: https://frida-ai.fr
- contact: tofmuck@frida-ai.fr
Ce depot est distribue sous licence MIT. Voir LICENSE.