justsay

Agentic System, written in Go.

Install

Install the latest published release with Go available:

curl -fsSL https://raw.githubusercontent.com/dhakad22klx/justsay/main/install.sh | sh
justsay

If the installer prints a PATH command, run it in your current shell or open a new terminal before starting justsay. See INSTALL.md for supported platforms, Gemini configuration, verification, updates, and removal.

Run from source

1. Clone the repository and change into its directory:

   git clone https://github.com/dhakad22klx/justsay.git
   cd justsay

2. Install the Go version declared in go.mod (currently Go 1.26.5), then confirm it is available:

   go version

3. Create a local environment file and fill in the required values. .env is read by the application and should not be committed.

   cp .env.example .env

   At a minimum, set:

   GEMINI_API_KEY="your Gemini API key"
   GEMINI_MODEL="your Gemini model ID"
   MOCK_AGENT_CALL="false"

   Set MOCK_AGENT_CALL to true to start without making model requests. Set HITL_ENABLED to true to hold the tool calls listed in agent/human-in-the-loop/hitl_config.yml for human approval. Both inmemory and redis support approvals, including those sent through a paired Telegram account. The inmemory backend requires no Redis configuration and keeps state only while the agent is running. The redis backend requires Redis configuration and supports persistent state shared across instances. See Agent state storage for configuration details.

4. Run the test suite from the repository root:

   go test ./...

5. Start the CLI from the repository root:

   go run .

Agent state storage

Select a backend in config.yml in the repository root. The bundled file selects inmemory:

state:
  backend: inmemory

To select Redis, change that file to:

state:
  backend: redis

To select through .env instead, remove state.backend from the YAML file (or remove the file), then set HITL_STATE_STORE="inmemory" or HITL_STATE_STORE="redis" in .env. A non-empty state.backend overrides .env, so changing .env alone does not override the bundled configuration.

If neither is set, the backend defaults to inmemory. Select redis explicitly to use Redis. Unknown backends cause an error, including an invalid HITL_STATE_STORE when YAML overrides it. The agent opens its store on the first pause or approval and retains it until shutdown; restart to apply configuration changes.

The inmemory backend requires no Redis connection or configuration. State is local to the running agent and is lost when it closes or the process restarts. Use redis for persistent state or state shared across instances. Redis requires REDIS_ADDR in .env; REDIS_PASSWORD, REDIS_DB (a numeric database index), and REDIS_KEY_PREFIX configure authentication, database, and key namespace.

Both backends use HITL_STATE_TTL from .env (default 24h). Each save restarts the expiry; HITL_STATE_TTL="0" disables it. In-memory entries expire on read and are also cleaned up periodically during writes.

CLI commands

Process commands work with go run . setup, go run . --help, or the equivalent installed justsay commands. Setup saves agent settings in config.yml and secrets in the existing credentials.json, preserving integration credentials. Provider settings use the generic model_provider entry, with api_key and model saved together in credentials.json; config.yml stores only model_provider.name (gemini). Previous provider-specific entries are not used or migrated. When HITL is enabled, setup asks for inmemory or redis (default inmemory, or the saved backend on reconfiguration). Only redis prompts for connection details. Both files and session transcripts use the single internal.ConfigDir constant in internal/config.go. The current value .justsay stores them in the user's ~/.justsay/ directory, with transcripts in ~/.justsay/sessions/, regardless of where the CLI is launched. Other relative values also resolve under the user's home directory; absolute paths are used directly. The value . uses the working directory for local development. The directory is created on save or session start, and credentials are readable only by their owner. Existing project-local files are not moved automatically; run justsay setup to configure the selected location. Setup does not read, write, or move .env. Runtime prefers the model and API key in credentials.json, falling back to .env for either missing value. The model is never read from config.yml. Other agent settings retain their runtime .env fallback until setup values are saved.

justsay version and justsay update are skeleton commands that return dummy output; version tracking and binary updates will be implemented later.

On startup, the terminal shows a panel with the selected model and saved integration status: Gmail authorization, Telegram pairing, and GitHub setup (currently not configured). Status is read locally; it is not a live service health check. Type help for integration setup commands. Narrow terminals use a compact layout, and NO_COLOR disables panel colors.

Command	Description
help	Show built-in and integration commands.
reset	Clear the current conversation.
/on	Enable real model calls (MOCK_AGENT_CALL=false).
/off	Mock model calls (MOCK_AGENT_CALL=true).
/verify telegram	Connect a Telegram bot to the running agent.
/verify gmail	Authorize Gmail sending with Google OAuth 2.0.
exit	Close the CLI.

Integrations

Use /verify telegram to pair a Telegram bot or /verify gmail to authorize Gmail sending. See INTEGRATIONS.md for setup steps and integration details.

Contributing

See CONTRIBUTING.md for the development and pull request guidelines.