q15-agent

q15-agent is the core of the stack. It owns:

• Prompt assembly — identity files from /memory/core/, working memory, and conversation history are assembled into the system prompt each turn.
• Tool wiring — file operations (read, write, edit), shell command execution via q15-exec, web search, and embedding search are all agent-owned tools.
• Telegram I/O — the agent communicates with users through a Telegram bot via long-polling.
• Memory management — turns are persisted to /memory/history/, and the agent can update its own semantic and working memory.
• Rooted file operations — the agent has read/write access to /workspace, /memory, and /skills.

Model selection

The agent uses capability-aware model selection. Each model in the config declares capabilities:

• text
• tool_calling
• reasoning
• image_input

When a request comes in, the agent infers which capabilities are needed and filters out models that cannot satisfy the requirements. The remaining models are tried in the order specified in agent.models.

If agent.cognition.models is set, those models are used for background cognition jobs. When omitted, cognition inherits agent.models.

Memory layers

The agent’s memory is structured into layers:

Layer	Path	Auto-injected?
Core identity	/memory/core/	Yes, every turn
Working memory	/memory/working/WORKING_MEMORY.md	Yes, every turn
Semantic memory	/memory/semantic/	No (tool-fetched)
Episodic history	/memory/history/	No (replayed by the agent)
Cognition state	/memory/cognition/	No (system-owned)

See the Memory concept page for details.

Conversation history

Completed turns are stored as JSON files under /memory/history/turns/YYYY/MM/DD/. The format is schema_version: 2 with canonical ordered messages[].parts[] entries of type text, reasoning, tool_call, and tool_result.

On startup, existing history is eagerly upgraded in place. Unreadable turn files are moved to quarantine. Cross-provider fallback uses portable transcript data first.

Tools

The agent can be configured with:

• web_search — Brave Search integration (requires BRAVE_API_KEY)
• embeddings — Qdrant-backed semantic search (requires Qdrant URL and Gemini API key)

Both are optional. Omit them from the config to disable.