@mcoda/codali

@mcoda/codali

Standalone tool-runner adapter for mcoda. codali runs a tool loop that can edit the repo directly via built-in tools, with optional streaming output and pre-flight cost estimates.

Usage

Run with a task file:

codali run --workspace-root . --provider openai-compatible --model gpt-4o-mini --task tasks/work.txt

Or pass the task on stdin:

echo "Fix the failing test" | codali run --workspace-root . --provider ollama-remote --model llama3

Smart pipeline

Enable the multi-phase pipeline with --smart (librarian/architect/builder/critic):

codali run --smart --workspace-root . --provider openai-compatible --model gpt-4o-mini --task tasks/work.txt

You can also set CODALI_SMART=1 in the environment.

Providers

• openai-compatible (default base URL: https://api.openai.com/v1)
• ollama-remote (default base URL: http://127.0.0.1:11434)
• stub (test-only provider)

Config and env

codali merges config in this order:

• CLI args
• codali.config.json or .codalirc
• Environment variables

Common environment variables:

• CODALI_WORKSPACE_ROOT
• CODALI_PROVIDER
• CODALI_MODEL
• CODALI_API_KEY
• CODALI_BASE_URL
• CODALI_STREAMING_ENABLED
• CODALI_STREAMING_FLUSH_MS
• CODALI_COST_MAX_PER_RUN
• CODALI_COST_CHAR_PER_TOKEN
• CODALI_COST_PRICING_OVERRIDES
• CODALI_TOOLS_ENABLED
• CODALI_ALLOW_SHELL
• CODALI_SHELL_ALLOWLIST
• DOCDEX_HTTP_BASE_URL
• CODALI_DOCDEX_REPO_ID

Routing config (per phase, optional) lives in codali.config.json:

{
  "routing": {
    "librarian": { "agent": "<librarian-agent-slug>", "temperature": 0.1 },
    "architect": { "agent": "<architect-agent-slug>", "temperature": 0.4 },
    "builder": { "agent": "<builder-agent-slug>", "temperature": 0.2, "format": "json" },
    "critic": { "agent": "<critic-agent-slug>", "temperature": 0.1 },
    "interpreter": { "agent": "<interpreter-agent-slug>", "temperature": 0.1 }
  },
  "limits": {
    "maxRetries": 3
  }
}

To enforce a GBNF grammar with Ollama:

{
  "routing": {
    "builder": { "format": "gbnf", "grammar": "root ::= \"ok\"" }
  }
}

Routing env overrides:

• CODALI_PROVIDER_LIBRARIAN
• CODALI_PROVIDER_ARCHITECT
• CODALI_PROVIDER_BUILDER
• CODALI_PROVIDER_CRITIC
• CODALI_PROVIDER_INTERPRETER
• CODALI_MODEL_LIBRARIAN
• CODALI_MODEL_ARCHITECT
• CODALI_MODEL_BUILDER
• CODALI_MODEL_CRITIC
• CODALI_MODEL_INTERPRETER
• CODALI_FORMAT_LIBRARIAN
• CODALI_FORMAT_ARCHITECT
• CODALI_FORMAT_BUILDER
• CODALI_FORMAT_CRITIC
• CODALI_FORMAT_INTERPRETER
• CODALI_GRAMMAR_LIBRARIAN
• CODALI_GRAMMAR_ARCHITECT
• CODALI_GRAMMAR_BUILDER
• CODALI_GRAMMAR_CRITIC
• CODALI_GRAMMAR_INTERPRETER
• CODALI_LIMIT_MAX_RETRIES

Logs

codali writes JSONL logs under logs/codali/<runId>.jsonl, rooted in the global workspace folder (~/.mcoda/workspaces/<workspace>/).

Docdex integration

codali calls docdex over HTTP for search/snippets/graphs and uses MCP for symbols/AST/memory. Ensure docdexd is running and DOCDEX_HTTP_BASE_URL is set if needed.