Providers & cost

Ortim routes each agent role to its own LLM provider. Most production setups run DeepSeek-only for budget reasons, then route the Architect and Security Reviewer to Anthropic where judgement matters. No API key at hand? Use Ollama locally for zero API cost.

Recommended: the config wizard

ortim config init (0.9.4+) writes ~/.ortim/config.toml. No .env required, and it works from any directory:

ortim config init                # pick provider → model → key, once
ortim config show                # verify what's resolved, with the source per field
ortim config set-role architect --provider anthropic   # role override

Or set environment variables

Still fully supported — ~/.ortim/config.toml only fills gaps, it never overrides an env var:

# .env — minimal
DEEPSEEK_API_KEY=sk-...

# Hybrid — premium reasoning where it pays off
ANTHROPIC_API_KEY=sk-ant-...
ARCHITECT_PROVIDER=anthropic
SECURITY_REVIEWER_PROVIDER=anthropic

Per-invocation override

The highest-precedence option, useful for one-off runs:

ortim run --provider ollama --model qwen2.5-coder:7b
ortim demo --provider ollama          # try the demo with zero API keys

Resolution order: --provider flag → shell / .env env var → ~/.ortim/config.toml → hardcoded default.

What a run costs

Approximate costs on observed proof-point runs (a Turkish brief, 6–8 tasks, 80%+ first-attempt approval):

| Setup | Planning chain | Per task | |---|---|---| | DeepSeek-only | $0.02–0.05 | $0.02–0.04 | | Hybrid (Architect + SecRev on Anthropic) | $0.05–0.10 | $0.04–0.08 | | Ollama-only | $0.00 (local) | $0.00 (local) |

Throughput on Ollama depends on your hardware.

Supported providers

anthropic, deepseek, ollama (local), and any OpenAI-compatible endpoint. Babel translates a non-English brief into structured intent before any expensive call, so token waste in non-English flows is contained, and premium models stay reserved for the roles that need them.