Chat

devrecall chat is a REPL over your work memory. Ask questions like “what was that auth bug I fixed in February?” and get an answer grounded in your actual commits, threads, and meetings.

Start a session

devrecall chat

DevRecall Chat — ask anything about your work history.
Type /help for commands, /quit to exit.

> what did I work on last week?

Chat needs an LLM configured. See Configure.

What it can answer

Pattern	Example
Time-window summaries	”what did I do in Q1?”
Specific recall	”what was that auth bug I fixed in February?”
Person-scoped	”what did I discuss with Sarah last week?”
Decision recall	”what did we decide about caching?”
Project-scoped	”summarize my work on the payment rewrite”
Metrics	”how many PRs did I review last quarter?”

How the agent answers

Chat is an agent loop. The LLM is given a small catalogue of read-only tools over your local SQLite database and decides which to call:

• current_time — anchor relative dates (“yesterday”, “Q1”)
• list_activities / count_activities — filter by date, source, type, identity
• search_activities — FTS5 keyword search
• semantic_search_activities — vector search via local ONNX embeddings
• get_activity — fetch the full body of one activity
• list_summaries / get_summary — read pre-computed periodic summaries
• list_identities / resolve_person — look up people

The model can call multiple tools per question and chain them. You can see exactly what it called with /trace after an answer.

See How it works for the full architecture.

Chat commands

/help            Show available commands
/quit            Exit (also /exit)
/clear           Clear conversation history
/search <query>  Raw FTS5 keyword search (no LLM, just hits)
/trace           Show the tool calls the agent made for the last answer
/stats           Memory stats — activity count, date range
/sync            Force re-sync of every wired source

Conversation memory

Chat remembers within a session — follow-ups work:

> what did I work on last week?
[answer]

> tell me more about the payment thing
[knows what "payment thing" refers to]

History is ephemeral. It’s not written to disk and is dropped on /quit or /clear.

Without an LLM

Chat itself requires an LLM. But devrecall search "auth token" works without one — pure FTS5 keyword search over your activities, no generation.

Privacy in chat

When you use BYOK (OpenAI / Anthropic), the retrieved context is sent directly from your machine to the LLM provider — not through any DevRecall server. With local Ollama, nothing leaves your machine at all.

DevRecall’s relay is never in the path for chat or summarization.