Phase 4 routing decisions depend on knowing whether the SLM classifier is actually firing or whether the heuristic is silently doing all the work. Adds the instrumentation to make that observable. router.ClassifierSource enum (heuristic / slm / slm_fallback) is set on Task by every classifier: - HeuristicClassifier → ClassifierHeuristic - slm.Classifier → ClassifierSLM on success, ClassifierSLMFallback when the SLM call fails or returns unparseable output The source is plumbed through router.Outcome to QualityTracker, which now maintains per-source counters alongside the existing per-arm × task EMA scores. QualitySnapshot serializes both (classifier_counts is omitempty for back-compat with pre-feature quality.json files). lazyClassifier logs at INFO the first time it falls back to heuristic because the SLM hasn't booted yet — distinguishes operational fallback from an unconfigured-SLM run. slm.Manager.Start() now records elapsed-to-healthy and the main.go goroutine logs it as part of the "SLM ready" event. Confirms whether short-lived runs are racing the boot cycle. New `gnoma router stats` subcommand prints both tables (arm × task quality, classifier source breakdown) from quality.json with a Phase 4 trust hint when the data is too sparse or the SLM share is low. 6 new tests cover ClassifierSource string/enum, heuristic + SLM source propagation, QualityTracker counter round-trip, and back-compat restore from a legacy quality.json without classifier_counts.
gnoma
A provider-agnostic agentic coding assistant built in Go. gnoma routes tasks to the best available LLM — cloud or local — through a multi-armed bandit router, while tools, hooks, skills, MCP servers, and plugins keep it extensible. Named after the northern pygmy-owl (Glaucidium gnoma); agents are called elfs (elf owl).
Quickstart
# Install
go install somegit.dev/Owlibou/gnoma/cmd/gnoma@latest
# Or build from source
git clone https://somegit.dev/Owlibou/gnoma && cd gnoma
make build # binary at ./bin/gnoma
# Set at least one provider key
export ANTHROPIC_API_KEY=sk-ant-... # or OPENAI_API_KEY, MISTRAL_API_KEY, GEMINI_API_KEY
# Run
gnoma # interactive TUI
echo "list files" | gnoma # pipe mode
gnoma --provider ollama # use a local model
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"
Extensibility (M8)
gnoma supports hooks, skills, MCP servers, and plugins.
MCP Servers
Connect any MCP-compatible tool server:
[[mcp_servers]]
name = "git"
command = "mcp-server-git"
args = ["--repo", "."]
timeout = "30s"
# Replace a built-in tool with an MCP tool
[mcp_servers.replace_default]
exec = "bash" # MCP tool "exec" replaces gnoma's built-in "bash"
MCP tools appear as mcp__{server}__{tool} (e.g., mcp__git__status), or under the built-in name when using replace_default.
Skills
Drop markdown files into .gnoma/skills/ or ~/.config/gnoma/skills/:
/skillname # invoke a skill
/skills # list available skills
Hooks
Run shell commands on tool events:
[[hooks]]
name = "block-rm-rf"
event = "pre_tool_use"
type = "command"
exec = "bash-safety-check.sh"
tool_pattern = "bash*"
Plugins
Bundle skills, hooks, and MCP configs into installable plugins:
gnoma plugin install ./my-plugin # install from directory
gnoma plugin list # list installed plugins
Plugins are pinned by SHA-256 of their plugin.json on first load
(Trust-On-First-Use). A manifest that changes between runs is refused with a
clear error and a re-enrollment hint. See docs/plugins-trust.md
and ADR-003.
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.