Previous test used compressed 8k images (0.01MB) which was meaningless.
Now tests with actual large noise images that don't compress.
Realistic payload limits discovered:
- Anthropic: 6 x 3MB = ~18MB total (not 32MB as documented)
- OpenAI: 2 x 15MB = ~30MB total
- Gemini: 10 x 20MB = ~200MB total (very permissive)
- Mistral: 4 x 10MB = ~40MB total
- xAI: 1 x 20MB (strict request size limit)
- Groq: 5 x 5760px images (5 image + pixel limit)
- zAI: 2 x 15MB = ~30MB (50MB request limit)
- OpenRouter: 2 x 5MB = ~10MB total
Also fixed GEMINI_API_KEY env var (was GOOGLE_API_KEY).
Related to #120
- Add AgentToolUpdateCallback type and optional onUpdate callback to AgentTool.execute()
- Add tool_execution_update event with toolCallId, toolName, args, partialResult
- Normalize tool_execution_end to always use AgentToolResult (no more string fallback)
- Bash tool streams truncated rolling buffer output during execution
- ToolExecutionComponent shows last N lines when collapsed (not first N)
- Interactive mode handles tool_execution_update events
- Update RPC docs and ai/agent READMEs
fixes#44
- Auto-enable all models after /login via POST /models/{model}/policy
- Use openai-responses API for gpt-5/o3/o4 models (not accessible via completions)
- Normalize tool call IDs when switching between github-copilot models with different APIs
(fixes#198: openai-responses generates 450+ char IDs with special chars that break other models)
- Update README with streamlined GitHub Copilot docs
- add GitHub Copilot model discovery (env token fallback, headers,
compat) plus fallback list and quoted provider keys in generated map
- surface Copilot provider end-to-end (KnownProvider/default, env+OAuth
token refresh/save, enterprise base URL swap, available only when
creds/env exist)
- tweak interactive OAuth UI to render instruction text and prompt
placeholders
gpt-5.2-high took about 35 minutes. It had a lot of trouble with `npm
check` and went off on a "let's adjust every tsconfig" side quest.
Device code flow works, but the ai/scripts/generate-models.ts impl is
wrong as models from months ago are missing and only those deprecated
are accessible in the /models picker.
- Image component returns correct number of lines (rows) for TUI accounting
- Empty lines rendered first, then cursor moves up and image is drawn
- This clears the space the image occupies before rendering
- Add spacer before inline images in tool output
- Create ShowImagesSelectorComponent with borders like other selectors
- Use showSelector pattern for /show-images command
- Add terminal.showImages setting to settings-manager.ts
- Add /show-images slash command (only visible if terminal supports images)
- ToolExecutionComponent checks both terminal support and user setting
- Shows text fallback when inline images are disabled
- Allow branch selector to open with single user message (changed <= 1 to === 0 check)
- Support in-memory branching for --no-session mode (no files created)
- Add isEnabled() getter to SessionManager
- Update sessionFile getter to return null when sessions disabled
- Update SessionSwitchEvent types to allow null session files
- Add branching tests for single message and --no-session scenarios
fixes#163
The theme file watcher was keeping the Node.js process alive indefinitely
even in print mode where hot-reload is unnecessary. This fix adds an
enableWatcher parameter to initTheme() and setTheme() functions, and only
enables watchers in interactive mode.
- Modified initTheme() to accept enableWatcher parameter (default: false)
- Modified setTheme() to accept enableWatcher parameter (default: false)
- Updated main.ts to only enable watchers in interactive mode
- Updated InteractiveMode to enable watchers when changing themes
- Add agentLoopContinue() to pi-ai for resuming from existing context
- Add Agent.continue() method and transport.continue() interface
- Simplify AgentSession compaction to two cases: overflow (auto-retry) and threshold (no retry)
- Remove proactive mid-turn compaction abort
- Merge turn prefix summary into main summary
- Add isCompacting property to AgentSession and RPC state
- Block input during compaction in interactive mode
- Show compaction count on session resume
- Rename RPC.md to rpc.md for consistency
Related to #128
Phase 1: Updated compaction.ts
- findCutPoint now returns CutPointResult with isSplitTurn and turnStartIndex
- Can cut at user, assistant, or bashExecution messages (never tool results)
- Added turnPrefixSummary support for split turns (parallel summarization)
- estimateTokens helper for context size estimation
Phase 2: Updated session-manager.ts
- CompactionEntry now has optional turnPrefixSummary field
- loadSessionFromEntries injects both summaries when turn was split
Phase 3: Updated agent-session.ts
- Overflow detection via isContextOverflow after agent_end
- Proactive compaction check on turn_end before next LLM call
- _abortingForCompaction flag to skip saving aborted messages
- Auto-retry after overflow recovery or proactive compaction
- New event fields: reason (overflow/threshold), willRetry
Phase 4: Updated interactive-mode.ts
- Shows reason in compaction status (Context overflow detected...)
- Shows retry status after compaction
Tests updated for new CutPointResult return type.
- Add maxLines and maxBytes fields to TruncationResult to track actual limits used
- Update tool-execution.ts to use actual limits from truncation result
- Add fallbacks to DEFAULT_MAX_* for backward compatibility with old sessions
- Fix outdated comments that said 30KB when default is 50KB
- Move RPC files to modes/rpc/ directory
- Add properly typed RpcCommand and RpcResponse types
- Expose full AgentSession API via RPC commands:
- State: get_state
- Model: set_model, cycle_model, get_available_models
- Thinking: set_thinking_level, cycle_thinking_level
- Queue: set_queue_mode
- Compaction: compact, set_auto_compaction
- Bash: bash, abort_bash
- Session: get_session_stats, export_html, switch_session, branch, etc.
- Add RpcClient class for programmatic access
- Rewrite tests to use RpcClient instead of raw process spawning
- All commands support optional correlation ID for request/response matching
- Trigger auto-compaction after agent_end instead of during message_end
- Show CompactionComponent after auto-compaction (same as manual /compact)
- Fix cut point to include bash executions before kept user message
- Stop backward scan at compaction, assistant, user, or toolResult boundaries
- Add AgentSessionEvent type extending AgentEvent with auto_compaction_start/end
- Emit events when auto-compaction starts and completes
- TUI shows loader during auto-compaction with escape to cancel
- Rebuilds chat UI when auto-compaction succeeds
Mario Zechner