vikingowl
15345540f2
Operational task types (debug, review, refactor, test, explain) now gate before orchestration in the keyword cascade. Previously, prompts like "review the orchestration layer" or "refactor the pipeline dispatch" matched "orchestrat"/"dispatch" and misclassified as TaskOrchestration. Planning is also moved below the operational types. Expanded orchestration keywords to cover common intent that the original four keywords missed: "fan out", "subtask", "delegate to", "spawn elf". Adds regression tests for false-positive cases and positive tests for new keywords.
gnoma
Provider-agnostic agentic coding assistant in Go. Named after the northern pygmy-owl (Glaucidium gnoma). Agents are called elfs (elf owl).
Build
make build # ./bin/gnoma
make install # $GOPATH/bin/gnoma
Providers
Anthropic
export ANTHROPIC_API_KEY=sk-ant-...
./bin/gnoma --provider anthropic
./bin/gnoma --provider anthropic --model claude-opus-4-5-20251001
Integration tests hit the real API — keep a key in env:
go test -tags integration ./internal/provider/...
OpenAI
export OPENAI_API_KEY=sk-proj-...
./bin/gnoma --provider openai
./bin/gnoma --provider openai --model gpt-4o
Mistral
export MISTRAL_API_KEY=...
./bin/gnoma --provider mistral
Google (Gemini)
export GEMINI_API_KEY=AIza...
./bin/gnoma --provider google
./bin/gnoma --provider google --model gemini-2.0-flash
Ollama (local)
Start Ollama and pull a model, then:
./bin/gnoma --provider ollama --model gemma4:latest
./bin/gnoma --provider ollama --model qwen3:8b # default if --model omitted
Default endpoint: http://localhost:11434/v1. Override via config or env:
# .gnoma/config.toml
[provider]
default = "ollama"
model = "gemma4:latest"
[provider.endpoints]
ollama = "http://myhost:11434/v1"
llama.cpp (local)
Start the llama.cpp server:
llama-server --model /path/to/model.gguf --port 8080 --ctx-size 8192
Then:
./bin/gnoma --provider llamacpp
# model name is taken from the server's /v1/models response
Default endpoint: http://localhost:8080/v1. Override:
[provider.endpoints]
llamacpp = "http://localhost:9090/v1"
Session Persistence
Conversations are auto-saved to .gnoma/sessions/ after each completed turn. On a crash you lose at most the current in-flight turn; all previously completed turns are safe.
Resume a session
gnoma --resume # interactive session picker (↑↓ navigate, Enter load, Esc cancel)
gnoma --resume <id> # restore directly by ID
gnoma -r # shorthand
Inside the TUI:
/resume # open picker
/resume <id> # restore by ID
Incognito mode
gnoma --incognito # no session saved, no quality scores updated
Toggle at runtime with Ctrl+X.
Config
[session]
max_keep = 20 # how many sessions to retain per project (default: 20)
Sessions are stored per-project under .gnoma/sessions/<id>/. Quality scores (EMA routing data) are stored globally at ~/.config/gnoma/quality.json.
Config
Config is read in priority order:
~/.config/gnoma/config.toml— global.gnoma/config.toml— project-local (next togo.mod/.git)- Environment variables
Example .gnoma/config.toml:
[provider]
default = "anthropic"
model = "claude-sonnet-4-6"
[provider.api_keys]
anthropic = "${ANTHROPIC_API_KEY}"
[provider.endpoints]
ollama = "http://localhost:11434/v1"
llamacpp = "http://localhost:8080/v1"
[permission]
mode = "auto" # auto | accept_edits | bypass | deny | plan
Environment variable overrides: GNOMA_PROVIDER, GNOMA_MODEL.
Testing
make test # unit tests
make test-integration # integration tests (require real API keys)
make cover # coverage report → coverage.html
make lint # golangci-lint
make check # fmt + vet + lint + test
Integration tests are gated behind //go:build integration and skipped by default.