TripleF

Developer ehildt released tripleF, an open-source agentic AI chat workbench that runs free open models fully locally via Ollama or through Ollama Cloud, with no usage meter or proprietary lock-in. The project, described as early in development but already a complete chat experience, ships with Docker Compose setup, optional integrations for the YouTube Data API, Serper (2,500 free credits, then roughly €5 for 5k searches), Bright Data SERP, and EODHD financial data, plus a Qdrant-backed memory layer with auto-reflection flags such as MEMORY_PARTITION_REFLECT_AUTO. tripleF is built using what its author calls "context coding," an AI-assisted development approach in which generative tools accelerate boilerplate while architectural decisions and failure modes are reviewed and tested.

An agentic AI workbench, fully open source. Chat with free open models — running fully locally via Ollama, or on Ollama Cloud for extra headroom. No lock-in, no meter, no black box — the whole machine is inspectable, and yours. Early in development; already a complete chat experience with features some proprietary chats still miss. tripleF aims for the top of the open-source community. This project is developed using context coding — a disciplined AI-assisted https://github.com/ehildt/tripleF/blob/main/.wiki/3-ai-assisted-development.md paradigm where generative tools accelerate boilerplate and exploration while every architectural decision, interface contract, and failure mode is deliberately reviewed, tested, and owned. We believe velocity without ownership accelerates directly into technical insolvency. 1. Install Docker/Docker Compose https://docs.docker.com/compose/install/ 2. Clone Repository https://github.com/ehildt/tripleF 3. Create API keys optional but recommended : - Ollama API key — for cloud models. Create one at ollama.com/settings/keys https://ollama.com/settings/keys · auth docs https://docs.ollama.com/api/authentication - YouTube Data API key free — enable the API in the Google Cloud Console, then create a key under Credentials. Getting started https://developers.google.com/youtube/v3/getting-started · obtaining credentials https://developers.google.com/youtube/registering an application - Serper API key — sign up https://serper.dev/signup and copy the key from the dashboard 2,500 free credits; afterwards super cheap — ~€5 yields 5k searches that will almost last you a lifetime - Bright Data API key + SERP zone — enterprise-grade alternative to Serper, slightly more expensive. API key docs https://docs.brightdata.com/api-reference/authentication · SERP API quickstart https://docs.brightdata.com/scraping-automation/serp-api/quickstart - EODHD API key — financial data API, free tier + tiered subscriptions. Register https://eodhd.com/register and grab the token from the dashboard · quick start https://eodhd.com/financial-apis/quick-start-with-our-financial-data-apis 4. Create the env files from the examples and set the keys: - ./apps/server/.env.example → ./apps/server/.env OLLAMA API KEY , YOUTUBE API KEY , SERPER API KEY , BRIGHT DATA API KEY + zones, EODHD API KEY - ./apps/memory/.env.example → ./apps/memory/.env OLLAMA API KEY , plus QDRANT EMBED MODEL — a model must be set; each memory layer can have a dedicated one, optional — otherwise it reuses the model from the request 5. Memory auto-maintenance optional but recommended — enable the AUTO flags in ./apps/memory/.env : - MEMORY PARTITION REFLECT AUTO=true — auto-trigger reflection after a partition's consolidation sweep - MEMORY COGNITION REFLECT AUTO=true — auto-trigger reflection after a cognition profile job - MEMORY ENCYCLOPEDIA REFLECT AUTO=true — auto-trigger reflection after the encyclopedia classification job - and downstream: MEMORY CONVICTION AUTO=true , MEMORY CLUSTER AUTO=true Otherwise reflection never runs on its own — you would need to trigger it manually via a cron job or scheduler hitting POST /memory/reflect , POST /memory/cognition/reflect , or POST /encyclopedia/reflect . All of these can also be toggled live in Settings settings → memory → configuration without a restart. 6. Run pnpm install 7. CD into ./apps/server and run pnpm db:push 8. Run docker compose -f infra.compose.yml up -d --remove-orphans 9. Run docker compose up -d --remove-orphans Contributions are welcome Please read CONTRIBUTING.md https://github.com/ehildt/tripleF/blob/main/CONTRIBUTING.md before opening an issue or pull request. Found a vulnerability? See SECURITY.md https://github.com/ehildt/tripleF/blob/main/SECURITY.md for our reporting policy. Community guidelines are in CODE OF CONDUCT.md https://github.com/ehildt/tripleF/blob/main/CODE OF CONDUCT.md . Distributed under the MIT License. See LICENSE https://github.com/ehildt/tripleF/blob/main/LICENSE for more information.