Blog
Technical updates from the Talon project. New features, architectural decisions, and the occasional opinion.
July updates: reasoning controls and safer provider sessions
Section titled “July updates: reasoning controls and safer provider sessions”2026-07-13
Personas can choose their reasoning effort, stateful Responses sessions stay scoped to the right model, and context rotation handles tool-heavy runs more accurately.
Talon + Postgram: persistent agent memory in one command
Section titled “Talon + Postgram: persistent agent memory in one command”2026-05-23
A zero-build starter stack wires Talon to Postgram for cross-conversation memory — three published images, one bootstrap script, and your agent remembers what you told it last week.
Talon on Docker: one command to a running daemon
Section titled “Talon on Docker: one command to a running daemon”2026-05-23
Docker v2 is a complete containerization story — compose, health checks, baked CLIs, and predictable permissions. One command to a running daemon.
Observational Memory: Durable Decision Logs Across Context Rotations
Section titled “Observational Memory: Durable Decision Logs Across Context Rotations”2026-04-08
Talon’s new session-observer mode replaces one-shot summaries with an append-only decision log. Each rotation preserves what happened, why it mattered, and what to do next.
Connect any OpenAI-compatible provider to Talon
Section titled “Connect any OpenAI-compatible provider to Talon”2026-04-08
Talon now supports a fourth execution provider for Ollama and other OpenAI-compatible endpoints, plus a new observational-memory mode for long-running threads.
April updates: CLI rewrite, Slack multi-bot routing, and scheduler fixes
Section titled “April updates: CLI rewrite, Slack multi-bot routing, and scheduler fixes”2026-04-03
CLI reference rewrite, durable scheduling clarifications, Slack shared-channel filtering, faster context rotation delivery, and a few cleanup fixes around background providers and memory grooming.
spark-coder — Fast Code Generation via Structured Output
Section titled “spark-coder — Fast Code Generation via Structured Output”2026-03-29
A new built-in sub-agent that uses gpt-5.4-spark to generate file operations from a task description. Also introduces requiresEnv — a generic mechanism for environment-gated sub-agent loading.
2026-03-29
A2A protocol, persona tools, queue fixes, and a full docs site. Quick changelog of everything that landed in the past two days.
persona_send and persona_list — A Practical Guide
Section titled “persona_send and persona_list — A Practical Guide”2026-03-29
How to configure and use the two new host tools that enable agent-to-agent delegation. Config examples, call patterns, and what happens when await_reply: true.
A2A — Agent-to-Agent Communication in Talon
Section titled “A2A — Agent-to-Agent Communication in Talon”2026-03-28
Talon now has a durable A2A layer. Personas can delegate tasks to each other over an internal queue, with hop enforcement and persistence across restarts.