Q
Which model for daily work?
The largest tool-calling local model that stays responsive on your hardware. Reserve frontier APIs for refactors that earn the spend.
Docs · Providers & Models
Providers & Models.
Local default, frontier on demand.
Direct answer
Three transports: Ollama (default, local, no account or key), generic OpenAI-compatible (OpenAI, Siemens preset, vLLM, LiteLLM, LM Studio, OpenRouter, custom gateways), and Anthropic Messages. Cloud providers are BYOK. Global config at ~/.anvaya/config.json, per-project overrides in .anvaya/config.json, secrets in credentials.json plus environment. Memory is model-agnostic, switching providers never invalidates the graph.
Providers & Models
Providers
01Ollama · defaultEmbeddings, summarization, generation, all local. Devstral is the default model; any tool-calling coder with 32K+ context works. Fully offline after one pull, no account or key.
02Anthropic · escalateFrontier reasoning for the hardest refactors. Configure a key; per-task model pick keeps spend deliberate.
03OpenAI · interoperateGPT-family reasoning inside Anvaya's memory loop. Same graph, different engine.
04Custom gateways · enterpriseData-driven provider defs without rebuilding, corporate gateways (Siemens pattern included), self-hosted endpoints, per-team routing.
Questions
Asked About Providers & Models.
Q
Does switching models wipe memory?
No. The graph stores typed nodes, links, and utility, not model-specific embeddings. Swap freely.
Q
How do credentials work?
~/.anvaya/credentials.json plus env (OPENAI_API_KEY, ANTHROPIC_API_KEY, OLLAMA_ENDPOINT). Project config overrides global; secrets never live in repos.
Run Agents That Fit On Your Laptop.
25.6 MB median RSS. 25 agents ran in parallel on a Core 2 Duo with 4 GB RAM. Hundreds on your machine. Zero cloud required on the Ollama path.
Requires Rust/cargo to build from source. Linux and macOS today, Windows not yet supported. Pre-1.0, public beta. Pricing TBD.